Crow

Crow

Crow is a modular, agentic framework and MCP platform that integrates with the services and AI tools you already use — run on hardware you own, with local or cloud models. It brings persistent memory, research, agents, P2P sharing, and a full self-hosted app ecosystem into one private interface that your AI clients can reach natively.

Built on the open Model Context Protocol standard. Published by Maestro Press | Product Page | Docs

Use it two ways — or both at once

As an agentic framework, build and run your own agents (the Bot Builder) over email, Discord, and voice — local and operator-gated. As an MCP platform, connect Crow to the AI client you already pay for (Claude Code, Claude Desktop, Cursor, opencode) as a native MCP server.

It rides your existing subscription as a first-class connector — the supported extension point, not a third-party harness you run instead of your client.

Both halves share the same memory, projects, files, and capabilities. Build an agent that runs autonomously on email, and your AI client in Claude Code still reaches the same memories that agent built up. That is the design: one system of record, two modes of access.

Crow is open source — developers can build integrations, skills, tools, panels, and bundles for it.

Your AI, your devices, your data

Your data stays on infrastructure you control. Pair Crow with a local model (Ollama or any OpenAI-compatible endpoint on your own machine) and nothing leaves your network; connect a cloud assistant (Claude, ChatGPT, Gemini, and others) and only what you choose to send that provider goes out, on your terms. Either way, the system of record is yours.

Privacy is honest here, not absolute marketing. The choice between local and cloud is yours, model by model and use case by use case.…