codebase-context-mcp
Static codebase analysis as MCP tools — give AI coding agents a map of your repo instead of letting them burn half their tokens rediscovering it file by file.
README
codebase-context-mcp
Static codebase analysis as MCP tools — give AI coding agents a map of your repo instead of letting them burn half their tokens rediscovering it file by file.
One tool call returns: file/module overview, internal import graph, HTTP routes
(Express / Fastify / Koa, with file:line), and cross-stack edges — which frontend
fetch/axios call hits which backend route.
Why
AI coding agents entering an unfamiliar codebase spend a large share of their context window
on grep/read loops just to learn the structure — and repeat it every session. Static
analysis answers most of those questions in milliseconds, deterministically, offline. This
server packages that as Model Context Protocol tools any
MCP client (Claude Code, Cursor, Cline, ...) can call.
Quick start
git clone https://github.com/oh-namgyu/codebase-context-mcp && cd codebase-context-mcp
npm install
# as a CLI
npx codebase-context analyze /path/to/repo # markdown architecture doc
npx codebase-context analyze /path/to/repo -f mermaid # flowchart
npx codebase-context analyze /path/to/repo -f json # raw model
# as an MCP server (Claude Code)
claude mcp add codebase-context -- node /path/to/codebase-context-mcp/src/mcp.js
MCP tools
| Tool | What it answers |
|---|---|
analyze_repo |
"What does this codebase look like?" — full architecture map (markdown or mermaid) |
get_routes |
"What HTTP endpoints exist?" — method, path, file:line, framework |
find_api_callers |
"Who calls this API?" — frontend call sites matched to a route |
Example output (this section is real output for a small Express repo):
# Architecture
- Routes: 7
- Cross-stack edges: 2
- Internal import edges: 5
## Routes
- `GET /api/claims` — src/server.js:29 (express)
...
## Cross-stack edges (frontend call → backend route)
- public/app.js:12 → `GET /api/claims` (src/server.js:29)
What it detects (v0.1)
- Languages: JavaScript / TypeScript / JSX / TSX (Babel parser, error-tolerant)
- Import graph: ESM
import+ CJSrequire, relative specifiers resolved to repo files - Routes: Express / Fastify / Koa-router member calls (
app.get('/x', ...)) and Fastify'sroute({method, url})object form — only in files that actually import those frameworks - Call sites:
fetch('/x')(incl. template literals →:param) andaxios.get('/x') - Cross-stack matching: method + path segments, route
:paramsmatch any segment
Not yet: Next.js/NestJS conventions, non-JS languages, incremental caching. PRs welcome.
Configuration
| Env | Default | |
|---|---|---|
CCM_MAX_FILES |
5000 |
file cap per analysis (guards huge monorepos) |
No network access, no telemetry, nothing leaves your machine.
Development
npm test # analyzer fixtures + render snapshots + CLI e2e + MCP stdio round-trip
MIT — see LICENSE. Security policy: SECURITY.md.
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。