codemunch-pro MCP server
✓ VerifiedIndexes codebases using AST extraction and hybrid search to provide AI agents with surgical, token-efficient access to symbols and call graphs.
Also known as BigJai/codemunch-pro
- Stars
- 3
- Tools
- 15
- Downloads, 30 days
- 248
- Latest version
- 1.2.0
- Last commit
- Mar 2, 2026
- License
- MIT
What is codemunch-pro MCP?
codemunch-pro indexes codebases using AST extraction and hybrid search to provide AI agents with surgical, token-efficient access to symbols and call graphs. It exposes 15 tools, such as index_folder, index_repo, list_repos.
Summary generated from the project's README
About codemunch-pro MCP server
Providing AI agents with surgical, token-efficient access to codebases is the primary function of codemunch-pro. It uses tree-sitter AST extraction and hybrid search to help developers navigate large repositories without consuming excessive context window space. By utilizing byte-offset seeking for symbol retrieval, the server can extract specific functions from large files while saving up to 99% of tokens compared to reading entire files.
Developers can use this server locally via Claude Desktop or Cline by adding it to their MCP configuration, or run it as an HTTP server. It supports ten languages including Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, and Ruby. The system handles both local directories through incremental SHA-256 based indexing and remote repositories from GitHub or GitLab via tarball downloads.
Technical capabilities include hybrid search combining FTS5 and vector similarity, call graph traversal to identify callers and callees, and full-text content search. It manages data using per-repo SQLite databases that store symbols, call edges, and file hashes to ensure only changed files are re-parsed during updates.
What you can do with it
- 1Minimize token usage during code analysisAsk the AI to retrieve the source code of a specific function without reading the entire file.
- 2Perform impact analysis for refactoringIdentify every function that calls a specific method before changing its signature.
Written from the project's README. Every use case is checked against a quote from it.
How to install codemunch-pro MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall codemunch-pro MCP in Claude Code
$ claude mcp add codemunch-pro -- uvx codemunch-proInstall codemunch-pro MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"codemunch-pro": {
"command": "uvx",
"args": [
"codemunch-pro"
]
}
}
}Install codemunch-pro MCP in VS Code
.vscode/mcp.json
{
"servers": {
"codemunch-pro": {
"type": "stdio",
"command": "uvx",
"args": [
"codemunch-pro"
]
}
}
}Install codemunch-pro MCP in VS Code CLI
$ code --add-mcp '{"name":"codemunch-pro","type":"stdio","command":"uvx","args":["codemunch-pro"]}'Install codemunch-pro MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"codemunch-pro": {
"command": "uvx",
"args": [
"codemunch-pro"
]
}
}
}Install codemunch-pro MCP in Codex
$ codex mcp add codemunch-pro -- uvx codemunch-proInstall codemunch-pro MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.codemunch-pro] command = "uvx" args = ["codemunch-pro"]
Install codemunch-pro MCP in Gemini CLI
$ gemini mcp add -s user codemunch-pro uvx -- codemunch-proInstall codemunch-pro MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"codemunch-pro": {
"command": "uvx",
"args": [
"codemunch-pro"
]
}
}
}codemunch-pro MCP tools 15
Detected from the READMEindex_folderindex_repolist_reposinvalidate_cachefile_treefile_outlinerepo_outlineget_symbolget_symbolssearch_symbolssearch_textget_callees
| index_folder | — | |
| index_repo | — | |
| list_repos | — | |
| invalidate_cache | — | |
| file_tree | — | |
| file_outline | — | |
| repo_outline | — | |
| get_symbol | — | |
| get_symbols | — | |
| search_symbols | — | |
| search_text | — | |
| get_callees | — |
Show all 15 toolsShow fewer tools
get_callersdiff_symbolsdependency_map
| get_callers | — | |
| diff_symbols | — | |
| dependency_map | — |
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablebigjai/codemunch-pro is public and reachablegithub.com/bigjai/codemunch-pro ↗Pass
- Package names this repository as its sourcecodemunch-pro names bigjai/codemunch-pro as sourcepypi.org/project/codemunch-pro/ ↗Pass
- Build provenance (Sigstore attestation)No build attestation published for this releaseNot checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-03-02github.com/bigjai/codemunch-pro ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from codemunch-proPass
- No critical advisory in the package or its dependencies64 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model15 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubCodeMunch Pro
Intelligent code indexing MCP server. 15 tools, 10 languages, tree-sitter AST extraction, hybrid search (FTS5 + vector), call graphs, remote repo indexing, incremental indexing.
Save 99% of tokens — get exact function source via byte-offset seek instead of reading entire files.
Install
pip install codemunch-pro
Quick Start
Claude Desktop / Cline
Add to your MCP client config:
{
"mcpServers": {
"codemunch-pro": {
"command": "codemunch-pro"
}
}
}
HTTP Server
codemunch-pro --transport streamable-http --port 5002
15 MCP Tools
| Tool | Description |
|---|---|
index_folder |
Index a local directory (incremental, SHA-256 based) |
index_repo |
Index a GitHub/GitLab repo (tarball download, no git needed) |
list_repos |
List all indexed repositories with stats |
invalidate_cache |
Force re-index a repository |
file_tree |
Get directory tree with file counts |
file_outline |
List symbols in a single file |
repo_outline |
List all symbols in repo (summary) |
get_symbol |
Get full source of one symbol (O(1) byte seek) |
get_symbols |
Batch get multiple symbols |
search_symbols |
Hybrid search (FTS5 + vector RRF) |
search_text |
Full-text search in file contents |
get_callees |
What does this function call? |
get_callers |
Who calls this function? |
diff_symbols |
What changed since last index? (PR review) |
dependency_map |
What does this file depend on? What depends on it? |
10 Languages
Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, Ruby
All via tree-sitter-language-pack — zero compilation, pre-built binaries.
Key Features
O(1) Symbol Retrieval
Every symbol stores its byte offset and length. get_symbol seeks directly to the function source — no reading entire files. A 200-byte function from a 40KB file = 99.5% token savings.
Incremental Indexing
Files are hashed (SHA-256). Only changed files are re-parsed. Re-indexing a 10K file repo after changing one file takes milliseconds.
Hybrid Search (FTS5 + Vector)
Combines BM25 keyword matching with semantic vector similarity using Reciprocal Rank Fusion. Search "authentication middleware" and find auth_middleware, verify_token, and login_handler.
Call Graphs
Traces function calls through the AST. get_callees("main") shows what main calls. get_callers("authenticate") shows who calls authenticate. Supports depth traversal.
Remote Repo Indexing (v1.1)
Index any public GitHub or GitLab repo by URL — no git binary needed. Downloads the tarball via API, extracts, and indexes. Cached locally with SHA-based freshness checks. Supports private repos with auth tokens and sparse paths.
Full-Text Content Search
Search raw file contents — string literals, TODO comments, config values, error messages. Not just symbol names.
How It Works
- Parse — tree-sitter builds an AST for each source file
- Extract — Walk AST to find functions, classes, methods, types, interfaces
- Store — SQLite database per repo with FTS5 virtual tables
- Embed — FastEmbed (ONNX, CPU-only) generates 384-dim vectors for semantic search
- Graph — Call expressions extracted from function bodies, edges stored and resolved
- Serve — FastMCP exposes 13 tools via stdio or HTTP
Architecture
~/.codemunch-pro/
├── myproject_a1b2c3d4e5f6.db # Per-repo SQLite database
├── otherproject_7890abcdef.db
└── ...
Each DB contains:
├── files # Indexed files with SHA-256 hashes
├── symbols # Functions, classes, methods, types
├── symbols_fts # FTS5 full-text search index
├── symbols_vec # sqlite-vec 384-dim vector index
├── call_edges # Call graph (caller → callee)
└── file_content_fts # Raw file content search
Use Cases
- AI Coding Agents: Give your agent surgical access to codebases without burning context
- Code Review: Find all callers of a function before changing its signature
- Onboarding: Search symbols semantically — "where is error handling?" finds relevant code
- Refactoring: Map call graphs before moving functions between modules
- Documentation: Extract all public APIs with signatures and docstrings
License
MIT
codemunch-pro MCP questions
How do I add codemunch-pro MCP to Claude Code?
Run claude mcp add codemunch-pro -- uvx codemunch-pro in your terminal. Add --scope user to make it available in every project.
How do I add codemunch-pro MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does codemunch-pro MCP provide?
It exposes 15 tools, including index_folder, index_repo, list_repos, invalidate_cache, file_tree.
How does it save tokens?
It uses byte-offset seeking to retrieve exact function source code instead of reading entire files, which can save up to 99% of tokens.
What programming languages are supported?
The server supports Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, and Ruby.
Can I index a remote repository?
Yes, you can index any public GitHub or GitLab repo by URL without needing a git binary installed.
Does it support semantic search?
Yes, it uses hybrid search combining BM25 keyword matching with semantic vector similarity via Reciprocal Rank Fusion.