home / mcp / jupytermcp server
JupyterMCP，允许通过MCP（Model Control Protocol）接口远程控制和操作Jupyter笔记本。该项目提供了一个sse协议的MCP，使AI助手或其他应用程序能够以编程方式创建、编辑和执行Jupyter笔记本单元格。
Configuration
View docs{
"mcpServers": {
"ilylty-jupytermcp": {
"url": "http://localhost:48080/sse"
}
}
}JupyterMCP is an MCP server that lets you remotely control and automate Jupyter notebooks using a dedicated MCP interface. With this server, you can open notebooks, run specific cells, insert or edit content, inspect notebook structure, and save or export results, all through a simple programmatic API.
You connect to the JupyterMCP server from an MCP client and issue commands to manage notebooks and cells. Common workflows include opening or creating a notebook, inserting new cells, executing single or multiple cells, retrieving outputs, and saving the notebook state. You can also configure slides mode for presentations and fetch structured information about the notebook and its cells.
A typical usage pattern looks like: open a notebook, insert a new code cell, write code, execute the cell, review the output, and save the notebook. You can also request information about all cells or the notebook’s basic metadata to build dynamic UI or automation routines.
Prerequisites you need before installing are Python 3.10 or later and either Jupyter Notebook or JupyterLab installed on your system.
Install and run the JupyterMCP server by following these steps in your terminal.
# Clone the project repository
git clone https://github.com/maskperson114514/jupyterMCP.git
cd jupyterMCP
# Install dependencies
pip install -r requirements.txt# Start the MCP server
python jupyterMCP.py --host 127.0.0.1 --port 48080Configure your MCP client to connect to the server via the SSE URL and enable the actions you want to perform on notebooks and cells. The following client configuration demonstrates enabling common actions and pointing to the local server over SSE.
{
"mcpServers": {
"jupyterMCP": {
"autoApprove": [
"save_notebook",
"get_cells_info",
"insert_and_execute_cell",
"run_cell",
"edit_cell_content",
"execute_cells_by_indices",
"run_all_cells",
"get_notebook_info",
"get_cell_text_output",
"set_slideshow_type"
],
"timeout": 60,
"url": "http://localhost:48080/sse",
"transportType": "sse"
}
}
}To run the server locally, start the Python script with the host and port you want to listen on. The example below binds to localhost on port 48080.
python jupyterMCP.py --host 127.0.0.1 --port 48080The server exposes an HTTP-based MCP endpoint at the configured URL for remote interaction. Clients use the SSE transport type to receive streaming updates and responses from operations such as opening notebooks, executing cells, and retrieving outputs. Use the provided client configuration to authorize and control which actions are automatically approved by default.
The JupyterMCP server provides a set of core functions you can call from your MCP client to manage notebooks and cells.
Limit access to the MCP endpoint to trusted clients and use authentication if available. Keep the server behind a secured network and consider enabling TLS/HTTPS if you expose the endpoint beyond a trusted environment.
If the server fails to start, confirm that Python 3.10+ is installed, dependencies are installed from requirements.txt, and the chosen host/port are not in use by another process. Check that the URL in client configurations matches the running server and that SSE transport is available on the server.
Opens an existing notebook or creates a new one if it does not exist. Returns the notebook reference and basic metadata.
Executes a single cell by its index within the active notebook and returns execution results.
Executes multiple cells in the specified order by their indices and aggregates outputs.
Saves the current state of the notebook to disk, preserving all cell contents and outputs.
Inserts a new cell at a specified position, fills it with code, and immediately executes it.
Inserts a new empty cell at a specified position without executing it.
Retrieves information about all cells in the current notebook, including type, content, and outputs.
Fetches basic metadata about the notebook, such as name, path, and structure.
Executes all code cells in the notebook in sequence and collects the complete output.
Returns the textual output produced by a specific cell.
Edits the content of a specified cell and applies changes to the notebook.
Sets the slideshow type for a cell to enable presentation mode.
Removes a specified cell from the notebook.