Trace Mcp
Code graph MCP: 81 languages, 91 frameworks, 72.7% fewer PR-review tokens, comprehension at parity
At a glance
Not tested yet- GitHub stars
- 204
- Weekly downloads
- 4,573
- Last commit
- 2026-11-19
- Licence
- MIT
- Transport
- stdio
- Auth
- none
npx -y trace-mcp
What it does
Trace Mcp is an MCP server that builds a dependency graph of a codebase and serves it to AI coding agents. It indexes source code and markdown knowledge vaults once, then answers questions from that precomputed structure instead of having the agent re-read files. It is for developers using MCP-compatible agents like Claude Code, Cursor, Codex, Windsurf, or Zed.

Tools it gives your agent
get_change_impact: returns the blast radius of a change across languages, with a risk score and linked architectural decisions.query_decisions: searches the decision knowledge graph linked to code.plan_turn: shows an optimal code subgraph, relevant past decisions, and dead-end warnings for a new task.search_sessions: searches past session content with FTS5.get_request_flow: traces a request from route through middleware, controller, service, and view.get_untested_symbols: classifies symbols as unreached or imported but never called in tests.get_subproject_impact: shows cross-subproject client calls with confidence scores.
Quick start
npm install -g trace-mcp trace init trace add
Add it to Claude Code from a terminal:
claude mcp add trace-mcp -- npx -y trace-mcp
Or add this to .mcp.json in your project root (Claude Code) or .cursor/mcp.json (Cursor):
{
"mcpServers": {
"trace-mcp": {
"command": "npx",
"args": [
"-y",
"trace-mcp"
]
}
}
}Environment variables
The registry entry declares no environment variables.
Example prompts
- What breaks if I change this model?
- Why was auth implemented this way?
- Show me the request flow from URL to rendered page.
- Find all untested code in this module.
What to watch out for
- The README reports a token reduction measured at trace-mcp 3.23.2 on 7 September 2026, and states this is a result from that build, not a claim about the current one.
- In a blind quality review of the same 60 pull requests, the trace-mcp arm understood the change in 67% of them against 65% for naive file loading, with 0.80 false positives per PR against 0.58.
- The Max tier's tweakcc pairing runs a third-party patcher for you, which the README describes as a declinable exception.
Not tested by tools4agents yet. Read the source repository before you install it.
Similar tools
Everything
The official test server. Exercises every part of the protocol, so it is the fastest way to check your setup works at all.
Filesystem
Gives an agent scoped read and write access to directories you name explicitly. The usual first MCP server people install.
Git
Read, search, and manipulate local Git repositories. Lets an agent inspect history and diffs without shelling out.
Sequential Thinking
Gives a model a scratchpad for multi step reasoning, letting it revise earlier steps instead of committing to a first answer.