Codemunch Pro

What codemunch-pro does

The codemunch-pro MCP server turns local codebases and public or private GitHub and GitLab repositories into searchable indexes. It extracts functions, classes, methods, types, and interfaces from Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, and Ruby source files.

Indexes are stored in a separate SQLite database for each repository. The index includes file metadata, symbols, full-text search data, vector representations, and call-graph edges. This lets an MCP client inspect a repository without repeatedly loading complete files into context.

The server is suited to coding agents, code review, onboarding, refactoring, and API documentation tasks. For example, an agent can locate an authentication implementation, retrieve one function's source, inspect its callers, and review symbols changed since the previous index.

How it works

The codemunch-pro MCP server parses supported files with tree-sitter and walks the resulting abstract syntax trees to identify code symbols and function calls. Each symbol records a byte offset and length, allowing direct retrieval from the source file rather than reading the whole file. The README describes this as O(1) symbol retrieval.

A repository's files are hashed with SHA-256. During later indexing passes, unchanged files are skipped and changed files are parsed again. Symbol search combines SQLite FTS5 keyword matching with semantic vectors generated by FastEmbed on the CPU, using reciprocal rank fusion to combine results. Call expressions are stored as graph edges and can be traversed to find callers or callees.

Remote repository indexing downloads GitHub or GitLab tarballs through an API, so a local Git installation is not required. Repositories are cached locally, and private repositories can be accessed with authentication tokens. Sparse paths are supported for remote indexing.

Setup and configuration

Install the package with:

pip install codemunch-pro

A stdio MCP client can start the executable directly:

{
  "mcpServers": {
    "codemunch-pro": {
      "command": "codemunch-pro"
    }
  }
}

The README gives Claude Desktop and Cline as client examples. An HTTP instance can be started on port 5002 with:

codemunch-pro --transport streamable-http --port 5002

The local data directory is shown as ~/.codemunch-pro/, with one database per repository. The provided material does not specify named environment variables or a detailed configuration file format.

Tools and capabilities

The codemunch-pro MCP server lists these operations:

• Index local folders incrementally or index GitHub and GitLab repositories.
• List indexed repositories, statistics, directory trees, and file counts.
• Force a repository re-index and inspect symbols at file or repository scope.
• Retrieve one symbol or multiple symbols by batch.
• Search symbol names and descriptions with hybrid search.
• Search raw file contents, including comments, literals, configuration values, and error text.
• Trace callers and callees, including depth traversal.
• Compare symbols since the previous index and map file dependencies.

The README labels the server as having 15 tools, while another implementation note mentions 13 tools; the tool table contains the detailed operation names.

Limitations and notes

Language support is limited to the 10 listed tree-sitter languages. Vector generation uses FastEmbed with ONNX on the CPU, so semantic search depends on that local indexing process. Remote indexing is described for GitHub and GitLab repositories, with tarball download and API-based access rather than Git.

Private repository access needs an authentication token, but the supplied material does not name the token variable or explain its exact client configuration. The documentation also does not state whether remote HTTP transport includes authentication controls. The project is released under the MIT license.

Getting started with this codemunch-pro MCP server

Always refer to the official documentation for the most accurate and up-to-date information.