Jupyter MCP Server
MCP server that connects AI assistants to Jupyter notebook environments for real-time notebook management. Provides tools to create, read, execute, and modify notebook cells, manage kernels, browse the Jupyter filesystem, and handle multimodal outputs including images and plots.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Jupyter notebook MCP. Executes arbitrary code — extreme caution. Token auth for Jupyter server. CRITICAL: code execution = full system access. Sandbox required for untrusted agent use.
⚡ Reliability
Best When
You have a JupyterLab instance running and want an AI assistant to directly create, execute, and iterate on notebook cells with full visibility into outputs including plots and images.
Avoid When
You need production workflow orchestration, are using Google Colab (support incomplete), or your AI client does not support multimodal responses.
Use Cases
- • AI-assisted data analysis and exploration in Jupyter notebooks
- • Executing code cells and iterating on results with AI guidance
- • Managing multiple notebooks and kernels through AI assistants
- • Generating and reviewing data visualizations with multimodal AI support
- • Automating repetitive notebook workflows like data cleaning pipelines
- • Teaching and learning with AI explaining and executing notebook code
Not For
- • Production data pipeline orchestration (use Airflow, Prefect, etc.)
- • Non-Jupyter notebook environments (Google Colab support still in development)
- • AI clients that cannot handle image/multimodal outputs
- • Environments without a running Jupyter server
Interface
Authentication
Token-based authentication via JUPYTER_TOKEN environment variable. Token is set when launching JupyterLab and passed to the MCP server configuration.
Pricing
Open source under BSD-3-Clause license. Requires a Jupyter server which is also open source.
Agent Metadata
Known Gotchas
- ⚠ Requires a running JupyterLab server with specific package versions (jupyterlab 4.4.1, jupyter-collaboration 4.0.2)
- ⚠ Specific pycrdt package conflicts require manual uninstall/reinstall of datalayer_pycrdt
- ⚠ JupyterHub and Google Colab support still in development
- ⚠ Multimodal output (images/plots) requires a multimodal-capable LLM and compatible MCP client
- ⚠ Docker configuration differs between macOS and Linux (host networking)
- ⚠ Multiple environment variables needed for configuration (JUPYTER_URL, JUPYTER_TOKEN, ALLOW_IMG_OUTPUT)
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Jupyter MCP Server.
Scores are editorial opinions as of 2026-03-06.