CocoIndex Code 🔌 MCP server Open source

What it does

CocoIndex Code (ccc) is a lightweight, Rust-powered semantic code search engine. Instead of grepping for exact strings, it builds an AST-based index of your repository so you — or your AI coding agent — can search it using natural-language descriptions of what you're looking for.

Tools it gives your agent

It gives an agent a fast, local "describe what you want, find the code" search primitive, aimed at cutting the number of tokens an agent burns exploring an unfamiliar codebase (the README claims roughly a 70% reduction versus plain text search). It supports 28+ languages including Python, JS/TS, Rust, Go, Java, C/C++, C#, and SQL.

Works with

The README documents integration with Claude Code (as a skill or over MCP), Cursor, Codex, OpenCode, Kilo Code, and a Grok plugin with hooks; it can also run standalone as an MCP server for other MCP-compatible agents.

How to install or connect

Install with pipx (recommended) or uv:

pipx install 'cocoindex-code[full]'

uv tool install --upgrade 'cocoindex-code[full]'

The [full] extra bundles local embeddings so no API key is required; a slim install instead relies on a cloud embedding provider. After installing:

ccc init
ccc index
ccc search "your query"

Maintenance and safety

Actively developed (2,700+ stars, 218 forks, 247 commits), Apache-2.0 licensed. Indexing and search run locally — your source code never leaves your machine. It does send anonymous usage telemetry (not code or queries) by default, disable with COCOINDEX_DISABLE_USAGE_TRACKING=1.

Who should use it

Developers who want their AI coding agent to search large or unfamiliar codebases faster and with fewer tokens, especially in Claude Code, Cursor, Codex or OpenCode workflows, without sending code to a third-party service.