marimo-team/marimo-pair
Drop agents inside running marimo notebook sessions
What it solves
marimo-pair allows AI agents to interact with and manipulate reactive Python notebooks. It transforms a marimo notebook into an environment where an agent can execute code and see the results in real-time, enabling a collaborative coding and data analysis environment between a human and an AI agent.
How it works
The project provides a set of tools (implemented as Agent Skills) that allow agents to discover running marimo servers and execute code within those notebooks. It integrates with the Agent Skills open standard and specifically provides a plugin for Claude Code, allowing the agent to interact with the notebook environment directly via bash scripts for discovery and execution.
Who it’s for
Developers and data scientists who use marimo reactive notebooks and want to AI agents (like Claude Code) to be able to write, run, and iterate on their code within that same environment.
Highlights
- Reactive Environment: Leverages marimo's reactivity to provide a dynamic environment for agents.
- Agent Skills Standard: Compatible with any agent that supports the Agent Skills open standard.
- ** uma Claude Code Integration**: Dedicated plugin for seamless integration with Claude Code.
- Server Discovery: Includes scripts to automatically discover running marimo servers.