Ctxo

<div align="center">

[![npm version](https://img.shields.io/npm/v/@ctxo/cli.svg)](https://www.npmjs.com/package/@ctxo/cli)
[![CI](https://github.com/alperhankendi/Ctxo/actions/workflows/ci.yml/badge.svg)](https://github.com/alperhankendi/Ctxo/actions/workflows/ci.yml)
[![Release](https://github.com/alperhankendi/Ctxo/actions/workflows/release.yml/badge.svg)](https://github.com/alperhankendi/Ctxo/actions/workflows/release.yml)

<picture>
  <source media="(prefers-color-scheme: dark)" srcset="docs/img/hero-svg.svg">
  <source media="(prefers-color-scheme: light)" srcset="docs/img/hero-svg.svg">
  <img alt="Ctxo — Code intelligence for AI agents" src="docs/img/hero-svg.svg" width="100%">
</picture>

</div>

***

### The Problem

AI coding assistants like Cursor, Claude Code, and Cursor rely on generic tools  `grep`, `find`, file reads to understand your codebase. On brownfield projects with thousands of files, this brute-force exploration creates a chain of problems:

* **Context window saturation** The agent fills its window reading files one by one, leaving little room for actual reasoning
* **Partial context hallucination** It sees a function but misses its dependencies, leading to wrong assumptions and broken suggestions
* **Lost-in-the-middle** Critical information buried deep in a long context gets ignored by the model
* **Context poisoning** Irrelevant code pulled in during exploration biases the model's output
* **Iteration overhead** Understanding one symbol takes 10-20 tool calls, each adding more noise to the context
* **Stale reasoning** After too many iterations, the agent contradicts its own earlier assumptions

The result: more tokens burned, slower responses, higher cost, and lower quality output.

### The Solution

Ctxo is an **MCP server** that **enhances** your existing AI tools with dependency-aware, history-enriched code intelligence. Instead of hundreds of `grep` and `read_file` calls, your agent gets the full picture symbol graphs, blast radius