Jupyter Notebook MCP Server

The Cursor Notebook MCP Server fills a critical gap for developers who rely on AI assistants to work directly inside Jupyter notebooks. Until version 0.50.5 of Cursor, agents in Agent mode were unable to edit notebook cells or files through the chat pane. This server resolves that limitation by exposing a rich set of MCP tools that let an AI model read, modify, and export notebooks in real time. Because it is built on the standard MCP framework, the same server can be configured for VS Code (Insiders), Claude Code, or any other MCP‑compatible client, making it a versatile bridge between conversational AI and notebook workflows.

At its core, the server uses nbformat to safely manipulate the JSON structure of files and nbconvert for exporting notebooks to Python scripts, HTML, Markdown, or other formats. All operations are scoped to user‑defined directories, ensuring that the agent cannot inadvertently alter files outside its sandbox. The API supports a wide range of actions—from creating new notebooks and adding or duplicating cells, to editing cell content, splitting long cells, and even updating cell outputs. Newer releases introduce advanced features such as SFTP support for remote SSH servers, streamable HTTP transport via FastMCP, and tools like that provide context about the notebook’s location on the server.

For developers, this means AI assistants can now act as true collaborators in data science and research projects. A model could, for example, read a dataset, suggest an exploratory analysis, automatically generate and insert code cells that perform the analysis, and export the final notebook as a Python script for deployment. In educational settings, an assistant could walk students through building a machine‑learning pipeline by editing notebook cells step by step. In research, it can help maintain reproducibility by ensuring that code and documentation stay synchronized within the same notebook environment.

Integration is straightforward: once the server is running, any MCP‑compatible client can request a tool by name. The server’s clean, declarative API ensures that the model receives only the data it needs, while the client handles authentication and routing. Because the tools are grouped under a single namespace, developers can easily discover and compose complex workflows—such as creating a notebook, populating it with a series of cells, exporting it to Python, and then pushing the result back to a version‑controlled repository—all through conversational commands.

What sets this server apart is its focus on security, consistency, and extensibility. By restricting file operations to preconfigured directories and validating notebook structures before each write, it guards against accidental corruption. The use of standard libraries (, ) guarantees compatibility with the broader Jupyter ecosystem. And its modular toolset—ranging from basic cell manipulation to bulk edits and output updates—provides a comprehensive toolkit that scales from simple scripts to complex, multi‑cell projects. This makes the Cursor Notebook MCP Server an indispensable component for any developer looking to embed AI intelligence directly into notebook‑centric workflows.