← Back to Discover

CocoIndex Code

A command-line tool for searching code by meaning. It builds an index from the syntax structure of your repository and can be used directly or by coding agents, through a skill or an MCP server.

MCP serversCommand-line tools
Visit website ↗

What it does

CocoIndex Code, run as the ccc command, indexes a repository using its abstract syntax tree and lets you search by describing what you are after, such as how user sessions are managed, instead of knowing exact names. It is built on CocoIndex, which the README describes as a Rust-based data transformation engine. There are three ways to use it. You can run ccc init, ccc index and ccc search yourself. You can install a ccc skill so a coding agent handles indexing and searching on its own, with setup shown for Claude Code and Grok plugins. Or you can run ccc mcp as an MCP server, with instructions for Claude Code, Codex, OpenCode, Kilo Code and Oh My Pi. The search tool can filter by language and path and supports pagination, and ccc grep matches code by structure without needing an index. The full install pulls in local embeddings that need no API key, while the slim install needs a cloud embedding provider and its key. Docker images come in a slim and a full variant. The README claims a 70% token saving, and the project is licensed under Apache License 2.0.

Capabilities

  • AST-based semantic search that takes natural-language queries
  • Works as a CLI, an agent skill or an MCP server
  • Structural code search with ccc grep, no index required
  • Local embeddings in the full install, or cloud providers through LiteLLM
  • Filters by language and path, with pagination
  • Slim and full Docker images

When to use it

  • Your coding agent spends a lot of context reading files just to find the right code
  • You need to find code by describing it in a large or unfamiliar repository
  • You want search that can run with local embeddings and no API key