Jupyter MCP Server is an extension for Jupyter environments that integrates LLMs with real-time coding sessions. By implementing the Model Context Protocol (MCP), it enables <\/span>AI<\/span><\/a> models to interact with Jupyter\u2019s kernel, file system, and terminal in a secure and context-aware manner. This blog will explore how you can use Jupyter MCP Server within your system.<\/span><\/p>\n MCP (Model Context Protocol) Servers are dedicated go-betweens that facilitate communication between AI assistants and applications or environments outside of them. They allow AI models to run in a state-aware manner, offering them real-time context such as variable values, code history, datasets, and results of execution.<\/span><\/p>\n Without MCP servers, AI models operate in a vacuum with no knowledge of pre-set variables. This isolation constrains their ability and potential for making mistakes.<\/span> MCP servers address this issue by providing AI with the power to reason, perform, and improve in a real-time environment, enabling them to become more functional, precise, and effective.<\/span><\/p>\n Jupyter MCP Server is a tool that acts as a bridge between a large language model <\/a>and the user\u2019s live Jupyter environment by primarily using the Model Context Protocol (MCP). It eliminates the limitations of models by removing the need for copying and pasting code and data. MCP has a secure protocol through which models can access and interact with various components of a Jupyter ecosystem. Thus, opening the concept of integrated, context-aware, and powerful AI-driven assistance.<\/span><\/p>\n Through the Model Context Protocol (MCP), the Jupyter <\/span>MCP<\/span><\/a> Server provides a structured way for external applications to interact with core Jupyter components.<\/span><\/p>\n Let\u2019s now understand some of these features in detail.<\/span><\/p>\n The system consists of three main components: kernel Interaction, File System Access, and Terminal Access. These components enable external applications to interface with Jupyter environments by running code, managing files, and controlling terminal sessions, respectively. In this section, we will understand each one of them in detail<\/span>:<\/p>\n It <\/span>lets Jupyter MCP Server run code inside the user\u2019s active kernels. It also inspects kernel state, retrieves execution results, and even manages kernel lifecycle.<\/span><\/p>\n How it works<\/b>: The MCP Client sends requests to the MCP API, specifying the target kernel and the action. Then, the MCP server communicates with Jupyter\u2019s kernel Manager to process the request.<\/span><\/p>\n Possible Actions<\/b>:<\/span><\/p>\n It provides controlled access to the user\u2019s workspace, allowing external applications to read, write, or manage files and directories.<\/span><\/p>\n How it works<\/b>: The MCP client requests file operations under the rules set by Jupyter\u2019s Content Manager and MCP\u2019s security policies.<\/span><\/p>\n Possible Actions<\/b>:<\/span><\/p>\n It<\/span> allows Jupyter MCP Server to interact with Jupyter\u2019s terminal sessions.<\/span><\/p>\n How it works<\/b>: The MCP client sends commands to a specific terminal session, and Jupyter\u2019s Terminal Manager processes the request, returning any output.<\/span><\/p>\n Possible Actions<\/b>:<\/span><\/p>\n This section outlines the steps required for the integration of the Jupyter MCP Server<\/strong>. But before diving into the installation and configuration process, let\u2019s first review the prerequisites needed to set up the Jupyter MCP Server<\/strong> properly<\/span>.<\/p>\n Once the prerequisites are met, we\u2019ll proceed with the installation and setup steps.<\/span><\/p>\n Since you\u2019re done with the prerequisites, let\u2019s perform the steps to download Jupyter MCP Server.<\/span><\/p>\n There are two ways in which you can download the Jupyter MCP server<\/span><\/p>\n Standard Installation<\/strong><\/p>\n Installing Jupyter MCP Server directly from PyPI using pip:<\/span><\/p>\n Development Installation\u00a0<\/strong><\/p>\n You can also clone the source code repository from <\/span>here<\/span><\/a><\/p>\n Step 1: Clone the above-mentioned repository<\/b><\/p>\n Step 2: Download Editable Mode<\/b>: This allows you to make changes in the source code that will be reflected.<\/span><\/p>\n Use the -e flag with pip to install the package in editable mode.<\/span><\/p>\n If you plan to run tests and contribute, you can use the following code:<\/span><\/p>\n Once installed (either the way explained above), you need to enable the extension for the Jupyter server. This loads and uses the MCP server functionality. Modifying Jupyter configurations, and updating the list of active extensions with the MCP server.<\/span><\/p>\n Once done with the above steps, you can verify your installation with the Jupyter server extension list.<\/span><\/p>\n Pro tip<\/strong>: If you see jupyter_mcp_server<\/em> in the list, then it\u2019s activated.<\/span><\/p>\n The Jupyter MCP Server won\u2019t be a visible interface instead, it is inside JupyterLAB or Notebook. It offers an HTTP API that other tools, such as Claude Desktop, AI models, backends, or plugins.<\/span><\/p>\n When using Claude Desktop, make changes in the <\/span>claude_desktop_config.json<\/span>. The TOKEN value and NOTEBOOK_PATH can be obtained from the terminal once you run your Jupyter notebook<\/span><\/p>\n Code for Windows:<\/strong><\/span><\/p>\n Code for Linux:<\/strong><\/span><\/p>\n Once approved, if it wants to interact with your Jupyter session, it sends a request to this API. The MCP Server handles the request by checking if it\u2019s allowed, communicating with the appropriate part of Jupyter, such as the kernel, file system, or terminal, and sending back the necessary response.<\/span><\/p>\n The API endpoint provided by the MCP extension, which lives at \/mcp\/v1, is just an added path to your Jupyter Server\u2019s base URL.<\/span><\/p>\n So, if your Jupyter server is running locally at <\/span>http:\/\/localhost:8888\/<\/span><\/a>, you can find the MCP API at <\/span>http:\/\/localhost:8888\/mcp\/v1<\/span><\/a>. This is where external tools like the Jupyter MCP server will reach out to communicate with your Jupyter environment.<\/span><\/p>\n HTTP requests are sent by the Jupyter MCP Server, like GET, POST, PUT, or DELETE, depending on<\/span> the task. Each request goes to a specific subpath under \/mcp\/v1. These request patterns and data structures make up what\u2019s known as the Model Context Protocol (MCP). For more details, you can refer to the main <\/span>README<\/span><\/a> file in the project explaining the role of the endpoint.<\/span><\/p>\n In the hands-on, we will see how to:<\/span><\/p>\n \nWhat are MCP Servers and why do we need MCP Servers?<\/h2>\n
![\"MCP](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/05\/233e7d1b-5749-4585-a884-12da7bf160bb.webp\")
<\/figure>\n<\/div>\nWhat is Jupyter MCP Server?<\/h2>\n
![\"Jupyter](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/05\/ChatGPT-Image-May-20-2025-05_16_48-PM.webp\")
<\/figure>\n<\/div>\nFeatures of Jupyter MCP Server<\/h3>\n
![\"Features\"](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/05\/ChatGPT-Image-May-20-2025-04_55_58-PM.webp\")
<\/figure>\n<\/div>\n\n
Key Functionalities of Jupyter MCP Server<\/h2>\n
Kernel Interaction<\/h3>\n
![\"Components\"](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/05\/ChatGPT-Image-May-20-2025-03_23_07-PM.webp\")
<\/figure>\n<\/div>\n\n
File System Access<\/h3>\n
\n
Terminal Access<\/h3>\n
\n
How to Integrate Jupyter MCP Server?<\/h2>\n
Prerequisites<\/span><\/h3>\n
\n
pip install jupyter-server<\/span><\/code><\/pre>\nInstallation<\/span><\/h3>\n
1. Download<\/span><\/h4>\n
pip install jupyter-mcp-server<\/span><\/code><\/pre>\ngit clone https:\/\/github.com\/datalayer\/jupyter-mcp-server.git<\/span>\ncd jupyter-mcp-server<\/span><\/code><\/pre>\nPip install -e<\/span><\/code><\/pre>\nPip install -e \u201c.[dev]\u201d\u00a0<\/span><\/code><\/pre>\n2.\u00a0 Activate Extension<\/span><\/h4>\n
jupyter server extension enable jupyter_mcp_server\u00a0\u00a0\u00a0\u00a0<\/span><\/code><\/pre>\njupyter server extension list<\/span><\/code><\/pre>\nJupyter MCP Server Working<\/h2>\n
{\n\n\u00a0\u00a0\"mcpServers\": {\n\n\u00a0\u00a0\u00a0\u00a0\"jupyter\": {\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"command\": \"docker\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"args\": [\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"run\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-i\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"--rm\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"SERVER_URL\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"TOKEN\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"NOTEBOOK_PATH\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"datalayer\/jupyter-mcp-server:latest\"\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0],\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"env\": {\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"SERVER_URL\": \"http:\/\/host.docker.internal:8888\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"TOKEN\": \"MY_TOKEN\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"NOTEBOOK_PATH\": \"notebook.ipynb\"\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0}\n\n\u00a0\u00a0\u00a0\u00a0}\n\n\u00a0\u00a0}\n\n}<\/code><\/pre>\nCLAUDE_CONFIG=${HOME}\/.config\/Claude\/claude_desktop_config.json\n\ncat $CLAUDE_CONFIG\n\n{\n\n\u00a0\u00a0\"mcpServers\": {\n\n\u00a0\u00a0\u00a0\u00a0\"jupyter\": {\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"command\": \"docker\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"args\": [\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"run\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-i\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"--rm\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"SERVER_URL\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"TOKEN\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"-e\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"NOTEBOOK_PATH\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"--network=host\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"datalayer\/jupyter-mcp-server:latest\"\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0],\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"env\": {\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"SERVER_URL\": \"http:\/\/localhost:8888\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"TOKEN\": \"MY_TOKEN\",\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\"NOTEBOOK_PATH\": \"notebook.ipynb\"\n\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0}\n\n\u00a0\u00a0\u00a0\u00a0}\n\n\u00a0\u00a0}\n\n}\n\nEOF\n\ncat $CLAUDE_CONFIG<\/code><\/pre>\nHands-on Application<\/h2>\n
\n