<h2>The Problem: Coding Agents Are Burning Tokens</h2>
If you’ve used Claude Code, Codex, Cursor, or any AI coding agent on a real codebase, you’ve probably noticed: they dump entire files into context just to understand your code structure. That’s expensive, slow, and wasteful.
The agent doesn’t need your whole file. It needs to know what functions exist, what classes are defined, and how they relate to each other.
<h2>The Solution: AST-Based Semantic Code Search</h2>
I built <a href="https://github.com/cocoindex-io/cocoindex-code" target="_blank" rel="noopener nofollow">cocoindex-code</a> - a super lightweight embedded MCP that:
<ul>
<li>Parses your code into ASTs using tree-sitter, extracting meaningful chunks (functions, classes, methods)</li>
<li>Creates semantic embeddings of those chunks</li>
<li>Lets your coding agent search by meaning, not just text matching</li>
<li>Only re-indexes changed files - built on a Rust-based incremental indexing engine</li>
</ul>
The result? 70% token savings and noticeably faster coding agent responses.
<h2>1-Minute Setup - No Config Needed</h2>
For Claude Code:
<pre><code>pipx install cocoindex-code
claude mcp add cocoindex-code -- cocoindex-code
</code></pre>
For Codex:
<pre><code>codex mcp add cocoindex-code -- cocoindex-code
</code></pre>
That’s it. No database, no API keys, no config files. It just works.
<h2>How It Works Under the Hood</h2>
<ol>
<li>Tree-sitter parsing breaks your code into semantic chunks (functions, classes, etc.) across 20+ languages</li>
<li>Local embedding model (SentenceTransformers) creates vector representations - completely free, no API key needed</li>
<li>SQLite + vector search stores everything locally and portably</li>
<li>Incremental indexing via <a href="https://github.com/cocoindex-io/cocoindex" target="_blank" rel="noopener nofollow">CocoIndex</a> (Rust engine) means only changed files get re-processed</li>
</ol>
When your agent needs to find code, it calls the <code>search</code> MCP tool with a natural language query and gets back exactly the relevant code chunks with file paths and line numbers.
<h2>Why It Went Viral</h2>
I think people resonated with a few things:
<ol>
<li>Real pain point - everyone using coding agents feels the token burn</li>
<li>Zero friction - one pip install and one MCP add command</li>
<li>No vendor lock-in - works with Claude, Codex, Cursor, or any MCP-compatible agent</li>
<li>Open source (Apache 2.0) - you can inspect every line of code</li>
<li>No API keys required - the default embedding model runs locally for free</li>
</ol>
<h2>Supported Languages</h2>
Python, JavaScript/TypeScript, Rust, Go, Java, C/C++, C#, Ruby, Kotlin, Swift, SQL, Shell, and more. It uses tree-sitter grammars so adding new languages is straightforward.
<h2>What’s Next</h2>
We’re actively working on:
<ul>
<li>Better embedding models optimized for code (try <code>nomic-ai/CodeRankEmbed</code> with a GPU)</li>
<li>Enterprise features for large codebases and shared indexing across teams</li>
<li>More MCP tools beyond search</li>
</ul>
The repo is at <a href="https://github.com/cocoindex-io/cocoindex-code" target="_blank" rel="noopener nofollow">github.com/cocoindex-io/cocoindex-code</a> - 520+ stars and growing fast.
Built with <a href="https://github.com/cocoindex-io/cocoindex" target="_blank" rel="noopener nofollow">CocoIndex</a>, our open-source Rust-based data indexing framework.
Would love to hear your experience if you try it out. Drop a comment or open an issue on GitHub!

Squad for devs building apps with GenAI and LLMs! Connect, share, exchange ideas, and get feedback. 

Build With GenAI

cocoindex-code is an open-source MCP server that reduces AI coding agent token usage by ~70% by parsing codebases into ASTs using tree-sitter, generating local semantic embeddings via SentenceTransformers, and storing them in SQLite for vector search. Instead of dumping entire files into context, agents query for relevant functions and classes by meaning. Setup requires just two commands with no API keys or config files. It supports 20+ languages, uses incremental Rust-based indexing to only reprocess changed files, and works with Claude Code, Codex, Cursor, and any MCP-compatible agent.

I Built a Code AST MCP That Saves 70% Tokens and Went Viral (54K+ Views)