# trace-mcp — Project Tracing MCP for Agents

> Index a project once, then reuse maps and task context via an MCP server. Includes a CLI to init and register projects for Claude Code, Cursor, and more.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Install / run:
   ```bash
   npm install -g trace-mcp
   ```
2. Start / smoke test:
   ```bash
   trace-mcp init && trace-mcp add
   ```
3. Verify:
   - Connect from your MCP client and ask for detected frameworks/project map, then request context for a small task.

## Intro

Index a project once, then reuse maps and task context via an MCP server. Includes a CLI to init and register projects for Claude Code, Cursor, and more.

- **Best for:** Teams who want fewer repeated “scan the repo again” passes and more deterministic context reuse across tasks
- **Works with:** Node.js CLI + MCP clients; codebases and markdown vaults (per repo docs)
- **Setup time:** 12 minutes

## How to Use It Well

- Setup time ~12 minutes (npm install + init + add a project)
- GitHub stars + forks (verified): see Source & Thanks
- CLI flow is 2 core commands (`init`, `add`) for first-time wiring

Treat it like a cache for agent reasoning: run the CLI once to index and register a project, then pull project maps, change impact, and task context through MCP tools instead of repeatedly loading the same files into the prompt.

### FAQ

**Q: What’s the big win?**
A: You stop re-deriving the same repo map for every task. The agent can pull structured project context when needed.

**Q: Do I have to keep files in my repo?**
A: The project describes keeping state under a home directory and keeping the repo clean unless you opt in.

**Q: Is it only for code?**
A: No. The docs describe indexing markdown vaults (notes, headings, links) as well.

## Source & Thanks

> Source: https://github.com/nikolai-vysotskyi/trace-mcp
> License: MIT
> GitHub stars: 69 · forks: 7

---

<!-- ZH -->

## 快速使用

1. 安装 / 运行：
   ```bash
   npm install -g trace-mcp
   ```
2. 启动 / 冒烟测试：
   ```bash
   trace-mcp init && trace-mcp add
   ```
3. 验证：
   - Connect from your MCP client and ask for detected frameworks/project map, then request context for a small task.

## 简介

将代码库/知识库索引为可复用的项目地图与任务上下文，通过 MCP 按需提供给 AI agent；提供 CLI 初始化并注册到 Claude Code、Cursor 等客户端，更适合长期复用。

- **适合谁：** 希望减少反复扫描仓库、让上下文可复用且更确定的团队
- **可搭配：** Node.js CLI + MCP 客户端；代码库与 Markdown 知识库（以仓库文档为准）
- **准备时间：** 12 分钟

## 实战建议

- 跑通约 12 分钟（npm 安装 + init + add 项目）
- GitHub stars + forks（已核验）：见「来源与感谢」
- 首次接入核心就两条命令（`init`、`add`）

把它当作 agent 推理的“结构化缓存”：先用 CLI 索引并注册项目，然后通过 MCP 工具按需拉取项目地图、变更影响与任务上下文，而不是每次都把同一批文件塞进 prompt。

### FAQ

**最大收益是什么？**
A: 把“仓库地图/上下文”一次性结构化，后续任务按需取用，减少反复扫描与 token 浪费。

**会往仓库里写很多东西吗？**
A: 仓库文档描述状态主要放在用户目录下；除非你选择启用忽略文件等可选配置。

**只适用于代码吗？**
A: 不止。文档还描述了对 Markdown 知识库的索引与引用（标题、链接、标签等）。

## 来源与感谢

> Source: https://github.com/nikolai-vysotskyi/trace-mcp
> License: MIT
> GitHub stars: 69 · forks: 7


---
Source: https://tokrepo.com/en/workflows/trace-mcp-project-tracing-mcp-for-agents
Author: MCP Hub