Contexo is an open-source MCP server that gives AI coding agents a persistent, shared memory of your project.
Instead of re-explaining the architecture, conventions, and past decisions every session (the "context tax"), agents pull what's already known before they start work and push back what they learn afterward.
Project knowledge lives in a git-backed repository, so it's versioned, diffable, reviewable, and shareable across teammates and tools. Rather than scattering context across chat histories, vendor-specific memory features, or local CLAUDE.md files, Contexo keeps decisions, gotchas, architecture notes, and the reasoning behind them in one place.
Because it's built on MCP, the same knowledge can be shared across Claude Code, Cursor, Codex, and other compatible agents.
Features include section-aware diffs, history, drift detection, agent-assisted merges, and structured capture of the why behind decisions—not just the final outcome.
Open-core: MIT-licensed Go CLI and server that you can self-host, plus an optional hosted version.
Server Config
{
"mcpServers": {
"contexo": {
"command": "ctx",
"args": [
"mcp"
]
}
}
}