Note: This repository is for organizing work on MCP support that will be contributed back to the mainline marimo project via pull request. We will import the marimo tree and work on it in a git worktree on a feature branch.

Development: The actual marimo codebase and implementation work can be found in the ./marimo/ worktree directory on the marimo/feature/mcp branch. See CLAUDE.md for repository structure and development workflow details.

This proposal outlines the integration of Model Context Protocol (MCP) support directly into marimo, enabling AI assistants and external tools to programmatically control marimo notebooks. By implementing both an HTTP-based MCP server (leveraging marimo's existing ASGI infrastructure) and a CLI-based MCP interface, we will unlock powerful AI-assisted workflows while maintaining marimo's core reactive principles.

ASGI Context: ASGI (Asynchronous Server Gateway Interface) is Python's standard for asynchronous web applications. marimo already uses ASGI to serve notebooks as web applications, making it a natural foundation for adding MCP server capabilities without introducing new dependencies.

1. Enable AI assistants (focus on Claude Code) to create, modify, and execute marimo notebooks programmatically
2. Provide seamless integration for both interactive (HTTP server) and batch (CLI) workflows
3. Maintain marimo's reactive execution model while exposing it to external control

1. HTTP-based MCP Server
2. CLI MCP Interface
3. Notebook Management
4. Reactive Notebook Editing

1. Developer Experience: Seamless AI assistance without leaving the marimo environment
2. Ecosystem Growth: First reactive notebook with native MCP support
3. Enterprise Ready: Secure, auditable AI-notebook interactions
4. Community Innovation: Enable new AI-powered notebook workflows

• Leverage existing ASGI infrastructure to minimize new dependencies
• Maintain backward compatibility with current marimo APIs
• Design for extensibility to support future MCP specification updates
• Ensure security through proper authentication and authorization
• Provide comprehensive documentation and examples

1. Full MCP protocol compliance for all implemented features
2. No performance degradation for non-MCP notebook usage
3. Successful integration with major AI platforms (Claude Desktop, Continue, etc.)
4. Positive community feedback and adoption
5. Clear documentation enabling third-party tool development

1. Community RFC to gather feedback on proposed features
2. Prototype implementation of core MCP server
3. Integration with marimo's existing test suite
4. Documentation and example creation
5. Beta testing with selected community members
6. Pull request submission with comprehensive testing

This native MCP integration will position marimo as the leading AI-native notebook platform, enabling workflows that aren't possible with traditional notebooks while maintaining the reactive execution model that makes marimo unique.