nteract 2.0.3a202603251613

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

nteract

An MCP (Model Context Protocol) server that connects AI assistants to Jupyter notebooks via the nteract desktop app.

Download the nteract desktop app — you'll need it to see notebooks, collaborate with agents, and manage environments.

Looking for the old Electron-based nteract desktop app? The source is archived at nteract/archived-desktop-app. The new native app is actively developed at nteract/desktop.

Bringing Agents in the Loop

We're in the preliminary 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
claude mcp add nteract -- uvx nteract

# Nightly
claude mcp add nteract-nightly -- uvx --prerelease allow nteract --nightly

Manual JSON config

{
  "mcpServers": {
    "nteract": {
      "command": "uvx",
      "args": ["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.

Available Tools

Tool	Description
list_active_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
set_cell	Update a cell's source and/or type
get_cell	Get a cell by ID with outputs
get_all_cells	View all cells in the notebook
replace_match	Targeted literal text find-and-replace in a cell
replace_regex	Regex-based find-and-replace in a cell
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
show_notebook	Open the notebook in the nteract desktop app (disabled with --no-show)
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

CLI Flags

Flag	Description
--version	Print version and exit
--nightly	Connect to the nightly daemon and open nightly app
--stable	Connect to the stable daemon and open stable app
--no-show	Do not register the show_notebook tool (for headless environments)

By default, nteract lets runtimed.default_socket_path() choose the socket. That means RUNTIMED_SOCKET_PATH wins when it is set; otherwise the package's build channel decides.

--stable and --nightly are mutually exclusive convenience overrides. When RUNTIMED_SOCKET_PATH is unset, they set it to runtimed.socket_path_for_channel(...). When RUNTIMED_SOCKET_PATH is already set, the explicit env var wins. In either case, the selected flag still controls which desktop app show_notebook tries to launch.

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