nteract 2.0.1a202603142058

Bring AI to Jupyter notebooks. MCP server for Claude, ChatGPT, Gemini, OpenCode and any agent.

nteract/nteract

What's going on?

If you're here looking for the Electron-based nteract desktop app, you can view the source in this repo. That desktop app is not actively maintained.

We're actively developing the spritual successor to the nteract desktop app in the nteract/desktop repo.

The New Desktop App

We're actively developing the spritual successor to the nteract desktop app in the nteract/desktop repo.

The new app is a native desktop app with instant startup and intelligent environment management.

Bringing Agents in the Loop

We're in the prelimiary stages of hooking up the realtime system from nteract/desktop to any agent of your choice. Collaborate with agents in notebooks, render interactive elements, and explore data together.

Quick Start

Claude Code

# Stable desktop 1.4.x / stable MCP
claude mcp add nteract -- uvx nteract

For desktop nightly / the upcoming 2.x transition, use the prerelease MCP package and nightly socket:

claude mcp add nteract-nightly -- env RUNTIMED_SOCKET_PATH="$HOME/Library/Caches/runt-nightly/runtimed.sock" uvx --prerelease allow nteract

That's it. Now Claude can execute Python code, create visualizations, and work with your data.

What is this?

nteract is an MCP (Model Context Protocol) server that connects AI assistants like Claude to Jupyter notebooks. It enables:

• Code execution: Run Python in a persistent kernel
• Real-time collaboration: Watch the AI work in the nteract desktop app
• Shared state: Multiple agents can work on the same notebook
• Environment management: Automatic Python environment setup

Example

Ask Claude:

"Help me visualize my log data"

Claude will:

1. Connect to a notebook session
2. Write and execute code
3. Generate visualizations
4. Show you the results

You can open the same notebook in the nteract desktop app to see changes in real-time and collaborate with the AI.

Installation

Stable line for desktop 1.4.x:

uvx nteract

Prerelease line for desktop nightly / 2.x transition:

uvx --prerelease allow nteract

Claude Code Setup

Add stable nteract as an MCP server:

claude mcp add nteract -- uvx nteract

Or manually add to your Claude configuration:

{
  "mcpServers": {
    "nteract": {
      "command": "uvx",
      "args": ["nteract"]
    }
  }
}

Using with Nightly

If you're using nteract desktop nightly builds, point at the nightly socket and allow prereleases:

claude mcp add nteract -- env RUNTIMED_SOCKET_PATH="$HOME/Library/Caches/runt-nightly/runtimed.sock" uvx --prerelease allow nteract

Release Tracks

• main publishes prerelease nteract builds for the 2.x transition and tracks runtimed 2.x prereleases.
• release/1.9.x is the stable maintenance line for desktop 1.4.x and stays on runtimed 1.9.0.

Available Tools

Tool	Description
list_notebooks	List all open notebook sessions
join_notebook	Join an existing notebook session by ID
open_notebook	Open an existing .ipynb file
create_notebook	Create a new notebook
save_notebook	Save notebook to disk as .ipynb file
create_cell	Add a cell to the notebook (use and_run=True to execute)
execute_cell	Run a specific cell (returns partial results after timeout)
run_all_cells	Queue all code cells for execution
append_source	Stream tokens into a cell (ideal for LLM output)
get_cell	Get a cell by ID with outputs
get_all_cells	View all cells in the notebook
set_cell_source	Replace a cell's entire source code
replace_match	Targeted literal text find-and-replace in a cell
replace_regex	Regex-based find-and-replace in a cell
set_cell_type	Change a cell's type (code, markdown, or raw)
move_cell	Reorder a cell within the notebook
clear_outputs	Clear a cell's outputs
delete_cell	Remove a cell from the notebook
interrupt_kernel	Interrupt the currently executing cell
restart_kernel	Restart kernel with updated dependencies
add_dependency	Add a Python package dependency
remove_dependency	Remove a dependency
get_dependencies	List current dependencies
sync_environment	Hot-install new deps without restart

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Claude    │────▶│   nteract   │────▶│   runtimed  │
│  (or other  │     │ MCP Server  │     │   daemon    │
│     AI)     │     │             │     │             │
└─────────────┘     └─────────────┘     └──────┬──────┘
                                               │
                    ┌─────────────┐             │
                    │   nteract   │◀────────────┘
                    │ Desktop App │  (real-time sync)
                    └─────────────┘

• nteract (this package): MCP server for AI assistants
• runtimed: Low-level daemon and Python bindings (docs)
• nteract desktop: Native app for humans to collaborate with AI

Real-time Collaboration

The magic of nteract is that AI and humans share the same notebook:

1. AI connects via MCP and runs code
2. Human opens the same notebook in nteract desktop
3. Changes sync instantly via CRDT
4. Both see the same kernel state

This enables workflows like:

• AI does initial analysis, human refines
• Human writes code, AI debugs errors
• Multiple AI agents collaborate on complex tasks

Development

# Clone
git clone https://github.com/nteract/desktop
cd desktop/python/nteract

# Install dependencies
uv sync

# Run tests
uv run pytest

Related Projects

• nteract/desktop - Native desktop app
• runtimed on PyPI - Low-level Python bindings

License

BSD-3-Clause