Jupyter MCP Server

Name: Jupyter MCP Server
Author: Datalayer

Read, edit and run Jupyter notebooks and their kernels from an MCP client

By Datalayer Python Remote (hosted) · Streamable HTTP Local · stdio Data and analytics Developer tools

GitHub PyPI Website

77 popularity score ★ 1.3k ↓ 56k/mo

Connect your client

Remote (hosted)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add --transport http jupyter https://mcp.datalayer.run/mcp

Claude (desktop and web)Settings → Connectors → Add custom connector

https://mcp.datalayer.run/mcp

ChatGPTSettings → Apps → Advanced → Developer mode, then Create app

https://mcp.datalayer.run/mcp

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "jupyter": {
      "url": "https://mcp.datalayer.run/mcp"
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "jupyter": {
      "type": "http",
      "url": "https://mcp.datalayer.run/mcp"
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.jupyter]
url = "https://mcp.datalayer.run/mcp"

Gemini CLITerminal

gemini mcp add --scope user --transport http jupyter https://mcp.datalayer.run/mcp

Local (PyPI)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add jupyter -- uvx jupyter-mcp-server@latest

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "jupyter": {
      "command": "uvx",
      "args": [
        "jupyter-mcp-server@latest"
      ]
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "jupyter": {
      "command": "uvx",
      "args": [
        "jupyter-mcp-server@latest"
      ]
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "jupyter": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "jupyter-mcp-server@latest"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.jupyter]
command = "uvx"
args = ["jupyter-mcp-server@latest"]

Gemini CLITerminal

gemini mcp add --scope user jupyter uvx -- jupyter-mcp-server@latest

Local (container)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add jupyter -- docker run -i --rm -e JUPYTER_URL -e JUPYTER_TOKEN -e ALLOW_IMG_OUTPUT datalayer/jupyter-mcp-server:latest

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "jupyter": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "JUPYTER_URL",
        "-e",
        "JUPYTER_TOKEN",
        "-e",
        "ALLOW_IMG_OUTPUT",
        "datalayer/jupyter-mcp-server:latest"
      ]
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "jupyter": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "JUPYTER_URL",
        "-e",
        "JUPYTER_TOKEN",
        "-e",
        "ALLOW_IMG_OUTPUT",
        "datalayer/jupyter-mcp-server:latest"
      ]
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "jupyter": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "JUPYTER_URL",
        "-e",
        "JUPYTER_TOKEN",
        "-e",
        "ALLOW_IMG_OUTPUT",
        "datalayer/jupyter-mcp-server:latest"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.jupyter]
command = "docker"
args = ["run", "-i", "--rm", "-e", "JUPYTER_URL", "-e", "JUPYTER_TOKEN", "-e", "ALLOW_IMG_OUTPUT", "datalayer/jupyter-mcp-server:latest"]

Gemini CLITerminal

gemini mcp add --scope user jupyter docker -- run -i --rm -e JUPYTER_URL -e JUPYTER_TOKEN -e ALLOW_IMG_OUTPUT datalayer/jupyter-mcp-server:latest

Settings 11

`JUPYTER_URL`	Optional: Address of the Jupyter server for both documents and kernels; defaults to http://localhost:8888 (use host.docker.internal from Docker on macOS and Windows).
`JUPYTER_TOKEN`	Optional: Token of that Jupyter server, used for documents and kernels alike.
`ALLOW_IMG_OUTPUT`	Optional: Return image outputs to the client; set false for models without image input.
`DOCUMENT_ID`	Optional: Notebook path, relative to where Jupyter started, to open by default.
`DOCUMENT_URL`	Optional: Separate address for notebook storage when it differs from the kernel server.
`DOCUMENT_TOKEN`	Optional: Token for the document server when it differs.
`CODE_SANDBOX_URL`	Optional: Separate address for code execution, such as a JupyterHub or Datalayer runtime.
`CODE_SANDBOX_TOKEN`	Optional: Token for the code execution server when it differs.
`SANDBOX_VARIANT`	Optional: Execution engine: jupyter-server (default), datalayer, kaggle, google-colab, modal, daytona, e2b, coreweave, cloudflare or monty.
`DOCUMENT_PROVIDER`	Optional: Where notebooks live: jupyter (default) or datalayer; replaces the deprecated PROVIDER.
`E2B_API_KEY`	Optional: API key when SANDBOX_VARIANT is e2b; other engines take their own keys such as DAYTONA_API_KEY or CWSANDBOX_API_KEY.

In a container, a setting goes in with -e NAME=value after docker run.

What it does

Jupyter MCP Server connects an MCP client to a running Jupyter server (JupyterLab, a plain Jupyter Server or JupyterHub) and works on its notebooks live, through Jupyter's real-time collaboration layer. The assistant can browse the server's files, open or create notebooks, read cells, insert, move, edit and delete them, clear outputs, restart kernels and execute cells or loose code, getting text, image and plot outputs back. Changes show up in an open JupyterLab tab as they happen.

Code normally runs on the Jupyter server's own kernels. A separate setting, SANDBOX_VARIANT, sends execution to another engine instead: Datalayer, Kaggle, Google Colab, Modal, Daytona, E2B, CoreWeave, Cloudflare or the local Monty interpreter, each with its own optional package and credentials. With the optional jupyter_mcp_sandboxes extension it also gains tools to launch, pick and stop such sandboxes. In JupyterLab mode (on by default) it can expose JupyterLab commands as extra tools, and it ships one prompt, jupyter_cite, for quoting cells into a conversation.

It is made by Datalayer, a company that sells hosted notebooks, and runs locally over stdio or as Streamable HTTP, or inside Jupyter Server as an extension. Datalayer also hosts it at https://mcp.datalayer.run/mcp with browser sign-in. Version 2 runs on the MCP Python SDK 2, so environments that pin mcp<2 must stay on the 1.x line; the old --provider option and PROVIDER variable are deprecated in favour of --document-provider and DOCUMENT_PROVIDER.

What it needs

To run it yourself	PythonorDocker

A running Jupyter server you can reach (JupyterLab with jupyter-collaboration, jupyter-mcp-tools and ipykernel installed for the quick start, or JupyterHub), its URL and token. Python with uv 0.6.14 or later for uvx, or Docker. Cloud sandbox engines need their extra package and an account or API key with that provider; the hosted endpoint needs a Datalayer account.

Limits

Real-time collaboration must be working in JupyterLab or the server cannot follow notebook changes. Sandbox lifecycle tools and every non-default sandbox engine need the separate jupyter_mcp_sandboxes extension. The code-sandboxes package version must match the server's (1.5.0 and later need code-sandboxes 1.1.1 or later), or execution fails with an unknown sandbox variant error. Execution times out after 120 seconds by default, capped at 3600. Set ALLOW_IMG_OUTPUT to false for models that cannot read images.

Who it suits

Suits data scientists and developers who already keep work in Jupyter and want an assistant to edit and run notebooks next to them, locally or on JupyterHub, with no account needed. It runs arbitrary code on whatever kernel or sandbox it points at, so it fits people who are comfortable granting that. Pick the hosted Datalayer endpoint if you want runs to continue after the client disconnects or need GPUs, accepting a vendor account. If you only need to read notebook files on disk without a live Jupyter server, a plain file-system server is lighter.

Tools 24

connect_to_jupyter: Switch to a Jupyter server at runtime with a given URL and token
list_files: List files and folders on the Jupyter server recursively
list_kernels: List the kernels available on the Jupyter server
use_notebook: Open or create a notebook and make it the active one
list_notebooks: Show the notebooks opened in this session
read_notebook: Return each cell's index, type, source and execution count
restart_notebook: Restart a notebook's kernel
unuse_notebook: Release a notebook and its resources
insert_cell: Add a cell at a given position in the active notebook
read_cell: Return one cell's content as text
edit_cell_source: Find and replace text inside one cell
overwrite_cell_source: Replace a cell's whole source
move_cell: Move a cell to another index
delete_cell: Remove cells, optionally returning their source
clear_cell_output: Wipe one code cell's outputs and execution count
execute_cell: Run a cell with a timeout and return its outputs
insert_execute_code_cell: Add a code cell and run it straight away
execute_code: Run code on the kernel or sandbox without saving it to the notebook
launch_sandbox: Start a code sandbox to use in place of a kernel (extension)
list_sandboxes: List the sandboxes launched so far (extension)
use_sandbox: Choose which sandbox execute_code runs on (extension)
terminate_sandbox: Stop a launched sandbox (extension)
notebook_run-all-cells: JupyterLab command exposed as a tool: run every cell (JupyterLab mode)
notebook_get-selected-cell: JupyterLab command exposed as a tool: get the cell selected in the UI (JupyterLab mode)

Popularity

77 popularity score, on our list (the bar is 36)

★ 1.3k ↓ 56k/mo

GitHub stars	1,320	37.5 points
Stars in 30 days	+21	6.7 points
PyPI downloads, last month	55,597	33.1 points

Measured 2026-11-07. How the score works.

Where it is published

Code	datalayer/jupyter-mcp-server ★ 1.3k
PyPI	jupyter-mcp-server 2.2.5, 2026-11-05, the one the command runs
Docker Hub	datalayer/jupyter-mcp-server
Hosted	`https://mcp.datalayer.run/mcp` Streamable HTTP · OAuth sign-in

What it can reach

Runs	On your machine, or on the maker's servers
Reaches	Files on your machine, Your code, The internet, Cloud resources
Changes things	Yes: it can write, send, create or delete
Sign-in	OAuth sign-in for the hosted endpoint; other to run it yourself
Licence	BSD-3-Clause (GitHub)
Last code change	2026-11-05

Runs on your machine (uvx or Docker) and talks to the Jupyter server you point it at with that server's token. Its tools can read, edit and delete notebook cells and run arbitrary code on the kernel, which has whatever file, network and shell access the kernel's host has. With a cloud SANDBOX_VARIANT, code runs on that provider's machines using your key there. The Streamable HTTP transport requires an --mcp-token unless started with --insecure-mcp-noauth. The hosted endpoint runs on Datalayer's servers and signs in with OAuth 2.1, issuing a token scoped to what you approve.

Over time

GitHub stars 1.3k +21 (+2 %)

1.3k1.3k

8 Oct 20267 Nov 2026

PyPI downloads a month 56k +18k (+49 %)

91k35k

12 May 20267 Nov 2026

Before the site measured it, read from pypistats' daily download counts.

Popularity score 77 +7 (+10 %)

7771

8 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-14.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![Jupyter MCP Server on EveryMCPServer](https://everymcpserver.com/badge/jupyter-mcp-server-2.svg)](https://everymcpserver.com/servers/jupyter-mcp-server-2)

Connect your client

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Settings 11

What it does

What it needs

Limits

Who it suits

Tools 24

Similar servers

dbt MCP Server

Optimizely Experimentation MCP Server

Vizro-MCP

CodeQL Development MCP Server

Convex MCP Server

Supabase MCP Server