M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
C

CodeMunch MCP

Community
AST-based code retrieval with 95%+ lower exploration token usage.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #199 of 438
Stars
★ 2.8k Very popular
Transport
stdio (local process)
Runtime
Python
Credentials
No credential needed
License
Other / unspecified
Last commit
Tools
13
54FMRS · D

A local-first MCP server for AI-assisted development, centered on tree-sitter symbol retrieval, structural code analysis, and token control. It is well suited to code-exploration-heavy workflows, while users should evaluate local index storage, optional telemetry, network-enabled extensions, and commercial licensing terms.

Strongest · Documentation 13/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
12/20
Maintenance
9/20
Documentation
13/20
Setup experience
12/20
Why each score
Reliability 8/20
The manifest declares a plausible PyPI stdio package and the launch path is understandable, but it says version 1.44.0 while the README references 1.56.0, 1.108.154, and 1.108.163, with inconsistent tool counts and capability descriptions. No verifiable CI, commit history, or key-path tests are supplied, so initialization, the tool-list handshake, and edge-error behavior cannot be established under the static calibration.
Security and permissions 12/20
The README discloses default anonymous savings telemetry, a background sender, session journaling of paths and queries, optional license/download/provider network calls, and opt-out settings; it also describes configuration, hooks, and login-service side effects. There is no evidence of credential theft or covert exfiltration, but default telemetry, potentially sensitive path/query logging, local indexing of source, and claimed access controls are not source-verified, while confirmation boundaries for side effects are incomplete, so points are deducted.
Maintenance 9/20
The README claims ongoing releases, upgrades, and a public repository, but those are unverified documentation claims. The manifest reports NOASSERTION, while the README's free-personal/commercial-paid dual-use terms lack verifiable license-file evidence. No dependency-update policy, security-response channel, commit activity, or governance evidence is supplied, preventing a high sustained-maintenance score.
Documentation 13/20
The material covers installation, client configuration, network behavior, optional dependencies, telemetry, troubleshooting, tool concepts, and cost controls in substantial breadth. Points are deducted for clear version and benchmark inconsistencies, highly promotional presentation, numerous unverified feature and test/telemetry claims, and incomplete supplied material at the configuration section.
Setup experience 12/20
It provides understandable pip and uvx paths plus examples for Claude Code, VS Code, Cursor, and other clients, including Windows PATH, Python/uv prerequisites, and debugging guidance. Points are deducted because the runtime and versions are not consistently pinned, clients require varying manual configuration, init may write configuration and hooks, optional provider/network dependencies add complexity, and no real CI or tests verify cross-platform connection setup.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesUses the network

Best for

  • Developers analyzing large or unfamiliar repositories
  • Engineering teams using Claude Code, Cursor, or another MCP client
  • Users who prioritize local indexing, targeted retrieval, and token-cost control

Not for

  • Users who only need general chat or web browsing
  • Environments that cannot or do not want to create local code indexes
  • Enterprise buyers requiring an official product-owner-maintained server or a clearly declared open-source license

Required permissions

  • Requires read access to local code directories or GitHub repositories being indexed.
  • Requires local write access for indexes and session data, stored by default under ~/.code-index/ and configurable with CODE_INDEX_PATH.
  • File watching, agent hooks, and login-service features require additional local configuration or service permissions when enabled.
  • Anonymous token-savings telemetry may be sent to the project's public meter; it can be disabled with share_savings or JCODEMUNCH_SHARE_SAVINGS=0.

Risks and side effects

  • Indexed source and session data are stored locally; the session journal records paths and query strings touched during a session.
  • Telemetry, AI summaries, embedding-model downloads, license validation, and Starter Pack operations can create outbound network requests when enabled or explicitly invoked.
  • Some optional extras can access a microphone or process images; these are not core MCP server functionality.
  • The README states that non-commercial personal use is free and commercial use requires a paid license; repository metadata lists the license as NOASSERTION.
  • Token savings and retrieval quality depend on index coverage, language support, and query quality and are not unconditional guarantees.

Setup

Before you start

Runtime：Python

Install the Python package with pip install jcodemunch-mcp, or run it directly with uvx jcodemunch-mcp. Add the supplied stdio configuration to an MCP client and restart the client. Alternatively, run jcodemunch-mcp init to detect and configure supported clients. Index a local directory or GitHub repository before querying it.

.mcp.json 
{"mcpServers":{"jcodemunch":{"command":"uvx","args":["jcodemunch-mcp"]}}}

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "jcodemunch": {
      "command": "uvx",
      "args": [
        "jcodemunch-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add jcodemunch -- uvx jcodemunch-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Run `jcodemunch-mcp --version` to confirm the install. In your client, ask the agent to answer via search_symbols and get_symbol_source; if it returns tens of precise lines instead of whole files, and get_session_stats reports tokens saved, the connection works.

Troubleshooting

  1. Verify that jcodemunch-mcp or uvx is installed and available on the MCP client's PATH.
  2. Run jcodemunch-mcp --version and jcodemunch-mcp config --check to validate installation, configuration, and prerequisites.
  3. Check the MCP client's server logs; Claude Code supports claude --mcp-debug.
  4. Confirm that the target directory has been indexed, and re-index after changes unless watch is enabled.
  5. If only a few tools are visible, inspect tool_profile, tool_surface, and disabled_tools; in counter mode, use menu() to browse the catalog.

Things to try

Once connected, you can ask your AI assistant things like:

  • Index this repo with jcodemunch.
  • Using jcodemunch, find the function that handles authentication and show me its source.
  • Run get_blast_radius on this function to show what breaks if I change it.
  • Use find_dead_code to check this repo for unreferenced functions.

Tools 13

search_symbols read-only
Search code symbols by name, kind, or language.
get_symbol_source read-only
Retrieve the exact source implementation of a symbol.
get_context_bundle read-only
Retrieve multiple related symbols and context within a token budget.
assemble_task_context read-only
Classify a natural-language task, extract anchor symbols, and assemble source-attributed context.
plan_turn read-only
Plan tool calls, target symbols, and a token budget before reading code.
get_ranked_context read-only
Return ranked symbol context within a token budget using relevance and PageRank.
find_importers read-only
Find code that imports a specified file or module.
get_blast_radius read-only
Analyze callers and dependencies potentially affected by changing a symbol.
Show 5 more tools
get_call_hierarchy read-only
Traverse callers and callees of a symbol.
find_dead_code read-only
Find unreachable symbols and files.
get_changed_symbols read-only
Map a Git diff to added, modified, or removed symbols.
search_ast read-only
Search structured AST patterns and preset anti-patterns across languages.
winnow_symbols read-only
Filter symbols using combined kind, complexity, call, file, Git-change, and importance criteria.

Use cases

Precise symbol retrieval from large codebases
Repository onboarding, debugging, and refactoring
Change-impact, call-hierarchy, and dead-code analysis
Reducing AI context usage and token costs during code exploration
Cross-language structural AST queries across more than 70 languages

Supported clients

Claude Code
Cursor
VS Code
CodexPartial support
ContinuePartial support
WindsurfPartial support
Autohand CodePartial support
Claude DesktopPartial support

Listed from the project's documentation, not tested by this site.

Overview

jcodemunch-mcp is an MCP server that parses local folders and GitHub repositories with tree-sitter and provides symbol-level code retrieval. It supports more than 70 languages and can return precise functions, classes, methods, constants, call relationships, and structural context instead of requiring agents to read entire files. Indexes are stored locally by default.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision bf118f74190e Data synced 2026-10-19 Read the FMRS scoring method