Cotext docs
Reference for Cotext — the local-first AI personalization extension and its CLI / MCP companions.
What Cotext is, in one paragraph
A browser extension that watches how you react to AI responses — likes, dislikes, free-form tags — and synthesizes a personal preference prompt that's injected into every AI you use (Claude, ChatGPT, Gemini, Grok, Mistral, Copilot). The synthesis runs on your machine (Chrome AI, WebGPU, or Ollama); nothing leaves unless you click Publish. The same prompt can be published as a content-addressed cotext.io/p/<hash> URL or pushed as a named @user/slug profile, and read by Claude Code / Codex via the MCP server.
Where to start
Pick the doc that matches what you're trying to do:
Guides (walkthroughs)
- Getting started — install the extension, build your AI personality, react your way to a synthesized prompt. Plain-English, no terminal required.
- For developers — CLI install, MCP integration with Claude Code / Codex, headless daemon mode, cloud-source push from the terminal.
- Self-hosting — run your own
cotext.iodeployment. Local dev, Postgres setup, Vercel deploy, custom categories.
Reference
- Architecture — components, data flow, the
.cotextfolder contract, source resolution. - CLI & MCP —
cotextandcotext-mcp, full command surface, every flag. - Telemetry — opt-in event schema and receiver recipes.
Other useful starting points
- Considering Cotext vs alternatives — Compare
- Want the source — github.com/cotext-io/cotext