trace-mcp
An MCP server that builds a code graph once and serves the answer to the agent instead of re-reading files
- 204 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads and indexes the whole project code
- Keeps a local background process running
Install
- Works in
Manual install
npm install -g trace-mcp
trace init
trace addGlobal install, trace init wires the server into the agent, trace add indexes the current repo.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads and indexes the whole project code
- Keeps a local background process running
This is third-party code. Review the repository files before installing.
What it does
trace-mcp indexes the repository into a framework-aware graph and keeps it fresh as the code changes. Instead of dozens of grep calls and file reads, the agent asks over MCP: what breaks if this changes, how a request flows, which symbols tests never reach. The server understands edges between layers, such as a controller wired to a view or a dependency created through injection. It runs locally, the code stays on your machine, and it installs as a standalone server or a plugin for Claude Code and Codex.
Who it is for. For developers who want to cut token spend and the agent's step count on large codebases.
Good fit when
- The agent re-reads the same files over and over in a large codebase
- You need the blast radius before changing a symbol or model
- You need to trace a request flow or find untested code
Not a fit when
- The project is small and the agent copes by reading directly
- You cannot run a local background indexing process
Example request
What breaks if I change this model, show the affected code through traceMCP
- Transport
- stdio
- Authentication
- not required
Limitations
It needs an installed server and local indexing, which costs time and memory on large repos. Some features belong to a paid tier and the desktop app, while the core MCP server and CLI are open under MIT. Suggestion accuracy depends on how well a given framework is supported.
How to disable. Remove the server from your agent's MCP config or uninstall the package: npm uninstall -g trace-mcp.
FAQ
Does my code leave for a third-party server?
No, the index is built and kept locally, and the server answers the agent from your machine.
Which agents does it work with?
Claude Code, Cursor, Codex, Windsurf and other MCP clients, with plugins for Claude Code and Codex.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that builds a code graph of functions, classes, calls and routes and answers the agent's structural queries locally
A local code graph with version history: an MCP server that tells the agent who calls a function and what an edit breaks
An MCP server that indexes code with tree-sitter and returns individual functions and classes to the agent instead of whole files
A transparent proxy between an AI client and an MCP server: shows every JSON-RPC frame in the terminal, checks a capture in CI and exports spans
A dependency-free CLI and MCP server that builds a call graph of a repository so the agent understands code without reading whole files
DevLens analyzes a repo once and serves agents ranked files, call flow and change impact over MCP, no file-by-file reading
README in short
The README argues that agents pay repeatedly for work already done: every turn they re-read files and re-traverse dependencies. trace-mcp builds a code graph once and serves it over MCP, so instead of many grep calls the agent gets an answer in a single request. It claims broad framework and language coverage, decision memory tied to symbols, and search across past sessions. Install is npm global, then trace init and trace add. A desktop app acts as a visual explorer over the same index.