CodeGraphMCPServer

A lightweight, high-performance source code analysis MCP server with zero configuration

🇯🇵 日本語版 README

Overview

CodeGraphMCPServer is an MCP server that understands codebase structure and provides GraphRAG (Graph Retrieval-Augmented Generation) capabilities. With a self-contained architecture requiring no external database, it enables structural understanding and efficient code completion from MCP-compatible AI tools (GitHub Copilot, Claude Desktop, Cursor, etc.).

🧠 GraphRAG Features

• Community Detection: Automatic code module clustering using Louvain algorithm
• LLM Integration: Multi-provider design supporting OpenAI/Anthropic/Local LLMs
• Global Search: Codebase-wide understanding using community summaries
• Local Search: Context retrieval from entity neighborhoods

✨ Features

Feature	Description
🚀 Zero Configuration	No external DB required, pip install && serve to start immediately
🌳 AST Analysis	Fast and accurate code analysis with Tree-sitter
🔗 Graph Construction	Builds graphs of relationships between code entities
🔍 14 MCP Tools	Dependency analysis, call tracing, code search
📚 4 MCP Resources	Entities, files, communities, statistics
💬 6 MCP Prompts	Code review, feature implementation, debug assistance
⚡ Fast Indexing	100K lines in under 30 seconds, incremental updates in under 2 seconds
🌐 Multi-language Support	Python, TypeScript, JavaScript, Rust, Go, Java, PHP, C#, C, C++, HCL, Ruby, Kotlin, Swift, Scala, Lua (16 languages)

Requirements

• Python 3.11+
• MCP-compatible client (GitHub Copilot, Claude Desktop, Cursor, Windsurf)

Installation

Install with pip

pip install codegraph-mcp-server

Install from source (for development)

git clone https://github.com/nahisaho/CodeGraphMCPServer.git
cd CodeGraphMCPServer
python -m venv .venv
source .venv/bin/activate  # Linux/macOS
pip install -e ".[dev]"

Quick Start

1. Index a Repository

# Full index
codegraph-mcp index /path/to/repository --full

# Incremental index (default)
codegraph-mcp index /path/to/repository

# Auto re-index with file watching (v0.7.0 NEW)
codegraph-mcp watch /path/to/repository
codegraph-mcp watch /path/to/repository --debounce 2.0  # 2 second debounce
codegraph-mcp watch /path/to/repository --community     # Community detection after re-index

Output example:

Indexed 16 entities, 37 relations in 0.81s

2. Check Statistics

codegraph-mcp stats /path/to/repository

Output example:

Repository Statistics
=====================
Repository: /path/to/repository

Entities: 16
Relations: 37
Communities: 0
Files: 1

Entities by type:
  - class: 2
  - function: 2
  - method: 11
  - module: 1

3. Search Code

codegraph-mcp query "Calculator" --repo /path/to/repository

4. Start as MCP Server

# stdio transport (default)
codegraph-mcp serve --repo /path/to/repository

# SSE transport
codegraph-mcp start --repo /path/to/repository --port 8080

MCP Client Configuration

Claude Desktop

~/.config/claude/claude_desktop_config.json:

{
  "mcpServers": {
    "codegraph": {
      "command": "codegraph-mcp",
      "args": ["serve", "--repo", "/path/to/your/project"]
    }
  }
}

Claude Code

# stdio transport
claude mcp add codegraph -- codegraph-mcp serve --repo /path/to/project

# HTTP transport (SSE server)
codegraph-mcp start --port 8080  # In another terminal
claude mcp add --transport http codegraph http://0.0.0.0:8080

VS Code (GitHub Copilot)

.vscode/settings.json:

{
  "mcp.servers": {
    "codegraph": {
      "command": "codegraph-mcp",
      "args": ["serve", "--repo", "${workspaceFolder}"]
    }
  }
}

Cursor

~/.cursor/mcp.json:

{
  "mcpServers": {
    "codegraph": {
      "command": "codegraph-mcp",
      "args": ["serve", "--repo", "/path/to/your/project"]
    }
  }
}

🛠 MCP Tools (14)

Graph Query Tools

Tool	Description	Main Arguments
query_codebase	Search code graph with natural language	query, max_results
find_dependencies	Find entity dependencies	entity_id, depth
find_callers	Find callers of function/method	entity_id
find_callees	Find callees of function/method	entity_id
find_implementations	Find interface implementations	entity_id
analyze_module_structure	Analyze module structure	file_path

Code Retrieval Tools

Tool	Description	Main Arguments
get_code_snippet	Get entity source code	entity_id, include_context
read_file_content	Get file content	file_path, start_line, end_line
get_file_structure	Get file structure overview	file_path

GraphRAG Tools

Tool	Description	Main Arguments
global_search	Cross-community global search	query
local_search	Local search in entity neighborhood	query, entity_id

Management Tools

Tool	Description	Main Arguments
suggest_refactoring	Suggest refactoring	entity_id, type
reindex_repository	Re-index repository	incremental
execute_shell_command	Execute shell command	command, timeout

📚 MCP Resources (4)

URI Pattern	Description
codegraph://entities/{id}	Entity details
codegraph://files/{path}	Entities in file
codegraph://communities/{id}	Community information
codegraph://stats	Graph statistics

💬 MCP Prompts (6)

Prompt	Description	Arguments
code_review	Perform code review	entity_id, focus_areas
explain_codebase	Explain codebase	scope, detail_level
implement_feature	Feature implementation guide	feature_description, constraints
debug_issue	Debug assistance	issue_description, context
refactor_guidance	Refactoring guide	entity_id, goal
test_generation	Test generation	entity_id, test_type

Usage Examples

Conversation with AI Assistant

You: What are the dependencies of the UserService class?

AI: [Using find_dependencies tool]
    UserService depends on:
    - DatabaseConnection (database.py)
    - Logger (utils/logging.py)
    - UserRepository (repositories/user.py)

You: What would be affected if I modify the authenticate method?

AI: [Using find_callers tool]
    Callers of authenticate:
    - LoginController.login() (controllers/auth.py:45)
    - APIMiddleware.verify_token() (middleware/api.py:23)
    - TestUserService.test_auth() (tests/test_user.py:78)

You: Explain the main components of this project

AI: [Using global_search tool]
    [Using explain_codebase prompt]
    
    This project uses a 3-tier architecture:
    1. Controllers layer: HTTP request handling
    2. Services layer: Business logic
    3. Repositories layer: Data access

Development

Run Tests

# Run all tests
pytest

# With coverage
pytest --cov=src/codegraph_mcp --cov-report=html

# Specific tests
pytest tests/unit/test_parser.py -v

Lint & Format

# Lint with Ruff
ruff check src tests

# Format with Ruff
ruff format src tests

# Type check with MyPy
mypy src

Architecture

src/codegraph_mcp/
├── __init__.py          # Package initialization
├── __main__.py          # CLI entry point
├── server.py            # MCP server
├── config.py            # Configuration management
├── core/                # Core logic
│   ├── parser.py        # Tree-sitter AST parser
│   ├── graph.py         # NetworkX graph engine
│   ├── indexer.py       # Repository indexer
│   ├── community.py     # Community detection (Louvain)
│   ├── semantic.py      # Semantic analysis
│   ├── llm.py           # LLM integration (OpenAI/Anthropic/Local)
│   └── graphrag.py      # GraphRAG search engine
├── storage/             # Storage layer
│   ├── sqlite.py        # SQLite persistence
│   ├── cache.py         # File cache
│   └── vectors.py       # Vector store
├── mcp/                 # MCP interface
│   ├── tools.py         # 14 MCP Tools
│   ├── resources.py     # 4 MCP Resources
│   └── prompts.py       # 6 MCP Prompts
└── languages/           # Language support (12 languages)
    ├── python.py        # Python extractor
    ├── typescript.py    # TypeScript extractor
    ├── javascript.py    # JavaScript extractor
    ├── rust.py          # Rust extractor
    ├── go.py            # Go extractor
    ├── java.py          # Java extractor
    ├── php.py           # PHP extractor
    ├── csharp.py        # C# extractor
    ├── c.py             # C extractor
    ├── cpp.py           # C++ extractor
    ├── hcl.py           # HCL (Terraform) extractor
    └── ruby.py          # Ruby extractor

Performance

Measured Values (v0.3.0)

Metric	Measured	Notes
Indexing speed	32 entities/sec	67 files, 941 entities
File processing speed	0.44 sec/file	Python/TS/Rust mixed
Incremental index	< 2 sec	Changed files only
Query response	< 2ms	Graph search

Target Values

Metric	Target
Initial index (100K lines)	< 30 sec
Incremental index	< 2 sec
Query response	< 500ms
Startup time	< 2 sec
Memory usage	< 500MB

License

MIT License - See LICENSE

Acknowledgments

• Model Context Protocol - MCP specification
• Tree-sitter - AST analysis
• NetworkX - Graph algorithms
• Microsoft GraphRAG - GraphRAG concept

Related Links

• MCP Python SDK
• MCP Specification