# nikolai-vysotskyi/trace-mcp [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/nikolai-vysotskyi/trace-mcp  
**GitHub Stars:** 198  
**npm Downloads (last month):** 2169  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/nikolai-vysotskyi-trace-mcp

## Description
Framework-aware code intelligence that indexes source code into a cross-language dependency graph. Understands framework semantics — routes, ORM relations, component rendering, DI trees — for navigation, impact analysis, call graphs, refactoring, security scanning, and cross-session memory.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "trace-mcp": {
    "command": "npx",
    "args": ["-y","trace-mcp"]
  }
}
```

## Documentation

## What nikolai-vysotskyi/trace-mcp MCP server does

nikolai-vysotskyi/trace-mcp MCP server builds an index of a source repository and represents relationships between symbols, files, and framework-level constructs as a dependency graph. The index spans multiple programming languages and is designed to retain framework meaning rather than treating the repository as plain text.

Supported semantic structures include application routes, ORM relationships, rendered components, and dependency-injection trees. This lets an agent investigate questions such as what may be affected by changing a model, where a symbol is used, or how a call path crosses framework boundaries.

The project is listed in the 💻 Developer Tools category. It is intended for agents working on local repositories, with the README stating that code stays local. A desktop application can also explore the same indexed graph through a GPU-based graph interface, separate from the MCP server workflow.

## How it works

The repository is indexed once and then queried through MCP. Rather than repeatedly loading files and reconstructing relationships during every task, an agent can request information from the precomputed graph. The index is kept incrementally fresh, so later requests can use updated repository structure without rebuilding the entire understanding from scratch.

One documented example is `get_change_impact`, which returns the likely blast radius of a change across areas such as source files, migrations, and dependency-injection code. The available capabilities cover code navigation, dependency and call-graph inspection, refactoring support, security scanning, and memory across sessions.

## Setup and configuration

Install the server globally with npm:

```bash
npm install -g trace-mcp
```

Run `trace init` once per machine to connect the server to an agent, then run `trace add` from the repository that should be indexed. The README also provides downloadable releases for macOS and Windows. No application installation is required for the MCP server itself.

## Tools and capabilities

The documented capability areas include:

- Navigate relationships in a codebase
- Inspect call graphs and dependency connections
- Calculate change impact with `get_change_impact`
- Analyze framework-aware structures such as routes and ORM relations
- Support refactoring and security-scanning workflows
- Preserve usable code context across agent sessions

The graph can be used by Claude Code, Cursor, Codex, Windsurf, Zed, and other MCP clients.

## Limitations and notes

nikolai-vysotskyi/trace-mcp MCP server is most relevant when the agent needs repository structure, dependencies, or framework semantics.

The published pull-request benchmark measured a specific build, version 3.23.2, on September 7, 2026. Its reported token and review results therefore describe that benchmark setup rather than guaranteeing the same results for every release or repository. The project is licensed under MIT.

The stated exception is a third-party patcher pairing available with the Max tier.

_Full upstream README: https://allmcps.com/mcp/nikolai-vysotskyi-trace-mcp/readme_