CodeAgent MCP
An autonomous code-reasoning agent that parses repositories using tree-sitter, indexes symbols into SQLite, and answers questions about the codebase through an agentic loop powered by Claude via OpenRouter.
README
CodeAgent MCP
An autonomous code-reasoning agent built on the Model Context Protocol (MCP). It parses your repository using tree-sitter, indexes all symbols into SQLite, and drives an agentic loop powered by Claude via OpenRouter to answer questions about your codebase.
Architecture
┌──────────────────────────────────────────────────┐
│ CLI (typer + rich) │
│ codeagent ask "what does Foo.bar() do?" │
└──────────────┬───────────────────────────────────┘
│
┌──────────────▼───────────────────────────────────┐
│ Agent Server (agent_server/) │
│ • Agentic loop — OpenRouter API (Claude) │
│ • Tool dispatch ↔ Code Server via MCP │
│ • Conversation history manager │
└──────────────┬───────────────────────────────────┘
│ MCP (stdio / SSE)
┌──────────────▼───────────────────────────────────┐
│ Code Server (code_server/) │
│ • FastMCP server │
│ • tree-sitter AST indexer → SQLite │
│ • Tools: search_symbols, find_callers, │
│ read_code, get_imports │
│ • watchdog file watcher (auto-reindex) │
└──────────────────────────────────────────────────┘
Setup Instructions
-
Clone the repository:
git clone <repo-url> && cd codeagent -
Set up virtual environment & install:
python3 -m venv .venv source .venv/bin/activate pip install -e . -
Configure environment variables:
cp .env.example .env # Open .env and set your OPENROUTER_API_KEY -
Index the repository:
codeagent index . -
Ask questions:
codeagent ask "How is the AST indexed into SQLite?"
Example Queries
- Find all functions:
codeagent ask "What functions are defined in this codebase?" - Trace function execution flow:
codeagent ask "Which components call start_watcher()?" - Explain architectural design:
codeagent ask "How does the file watching indexing system work?"
Environment Variables
| Variable | Required | Description |
|---|---|---|
OPENROUTER_API_KEY |
✅ | Your OpenRouter API key (sk-or-...) |
CODEAGENT_REPO |
optional | Default repo path (defaults to .) |
Tech Stack
| Layer | Library |
|---|---|
| LLM | openai SDK → OpenRouter (anthropic/claude-sonnet-4-5) |
| MCP | mcp (FastMCP) |
| Code parsing | tree-sitter ≥0.22 + tree-sitter-python + tree-sitter-javascript |
| Database | aiosqlite (async SQLite, no ORM) |
| File watching | watchdog |
| CLI | typer + rich |
| Testing | pytest + pytest-asyncio |
Development
# Run all tests
pytest -v
# Run only indexer tests
pytest tests/test_indexer.py -v
# Index a repo manually (Python API)
python3 -c "
import asyncio
from code_server.indexer import index_repo
asyncio.run(index_repo('/path/to/repo'))
"
Project Structure
codeagent/
├── code_server/
│ ├── indexer.py # tree-sitter AST → SQLite (Phase 1 ✅)
│ ├── tools.py # MCP tools: search_symbols, find_callers, read_code, get_imports
│ ├── server.py # FastMCP server
│ └── watcher.py # watchdog file watcher
├── agent_server/
│ ├── loop.py # agentic loop (OpenRouter + tool use)
│ └── history.py # conversation history manager
├── cli/
│ └── main.py # typer CLI
├── tests/
│ ├── test_indexer.py ✅ 6 passing
│ ├── test_tools.py
│ └── test_loop.py
├── .env.example
└── pyproject.toml
Build Phases
- [x] Phase 0 — Project scaffold,
pyproject.toml, venv, package installs - [x] Phase 1 — AST indexer (
indexer.py) - [x] Phase 2 — Code tools (
tools.py) - [x] Phase 3 — FastMCP server (
server.py) - [x] Phase 4 — Conversation history manager (
history.py) - [x] Phase 5 — Core Agentic loop (
loop.py) - [x] Phase 6 — Rich CLI (
main.py) - [x] Phase 7 — Watchdog file watcher (
watcher.py) - [x] Phase 8 — Verification & Test suite (36/36 tests passing)
- [x] Phase 9 — Complete README documentation
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。