Jupyter Claude Plugin

A JupyterLab extension enabling collaborative development between users and Claude. Chat about your notebook, generate cells from prompts, explain and document existing code, and fix errors — all within JupyterLab.

Status: early development, scaffold phase complete. See docs/PLAN.md for the implementation plan and docs/DECISIONS.md for the decision log.

Features (v1)

• Chat about the notebook — ask Claude to analyze, summarize, or answer questions about your entire notebook
• Generate a cell — describe what you want; choose cell type (code/markdown/raw); Claude generates and inserts it
• Explain a cell — Claude documents an existing cell with a markdown explanation
• Fix last error — Claude sees the error from your last cell run and offers a corrected version
• Runtime tier selector — switch between opus/sonnet/haiku in the chat panel for quality/speed tradeoffs (Bedrock only)
• MCP server management — enable/disable individual MCP servers from ~/.claude.json per session via dialog command (defaults to Jupyter only for faster startup)
• Notebook toolbar buttons — all five commands as icons in toolbar: Open Chat, Generate cell, Explain cell, Fix last error, and MCP server selector
• Visual feedback — animated "Claude is thinking…" spinner while processing; disabled tier selector during requests

Requirements

• JupyterLab >= 4.0.0
• Node >= 20
• Python >= 3.10 (this project targets 3.13)
• uv package manager (via Homebrew: brew install uv)

Install (from PyPI, recommended)

pip install --pre jupyter-claude-plugin

--pre is required for alpha releases (currently 0.1.0a1). Runtime dependencies (claude-agent-sdk, jupyter-mcp-server, jupyter-collaboration) are pulled in automatically. See docs/USER_GUIDE.md for backend configuration and usage.

Development install (from source)

cd /path/to/Jupyter-Claude_Plugin
uv venv
source .venv/bin/activate

jlpm install
uv pip install -e ".[dev,test]"
jlpm build
jupyter labextension develop . --overwrite
jupyter server extension enable jupyter_claude

jupyter lab

Global MCP servers (user scope, one-time)

claude mcp add --scope user filesystem -- npx -y @modelcontextprotocol/server-filesystem $HOME
claude mcp add --scope user memory -- npx -y @modelcontextprotocol/server-memory
claude mcp add --scope user sequential-thinking -- npx -y @modelcontextprotocol/server-sequential-thinking
claude mcp add --scope user fetch -- uvx mcp-server-fetch
claude mcp add --scope user git -- uvx mcp-server-git
claude mcp add --scope user jupyter --transport http http://localhost:8888/mcp

Backend

The extension supports two Claude backends. Pick one in the JupyterLab Settings Editor (or via jupyter_server_config.py) — everything else is derived from that choice.

Anthropic direct

Set ANTHROPIC_API_KEY in the environment before launching Jupyter. The extension will use the model trait (default claude-opus-4-8).

AWS Bedrock via shared SSO

Configuration lives in the extension traits (not shell env — the extension writes the right vars into the Claude Agent SDK subprocess itself):

• backend = "bedrock"
• aws_region = "us-east-1" (or your region)
• aws_profile = "<your profile>" — a named profile from ~/.aws/config (enables shared AWS SSO; profile and region are forwarded to the SDK subprocess)
• default_opus_model, default_sonnet_model, default_haiku_model — inference-profile model IDs (defaults: us.anthropic.claude-opus-4-7, us.anthropic.claude-sonnet-4-6, us.anthropic.claude-haiku-4-5-20251001-v1:0)

Refresh your SSO session and start Jupyter with aws-vault so the subprocess inherits temporary STS credentials without leaking them to disk:

aws sso login --profile <your profile>              # once per session
aws-vault exec <your profile> -- jupyter lab

If you prefer not to use aws-vault, AWS_PROFILE=<name> AWS_REGION=<region> jupyter lab also works — the extension forwards those to the SDK.

See docs/PLAN.md for design details.

Architecture

JupyterLab (browser)
  └─ Frontend extension (TS/React)
       ├─ Right-side chat panel
       ├─ Cell toolbar buttons
       └─ Context-menu items
                          ↓ WebSocket
                  Server extension (Python)
                    └─ ClaudeSDKClient
                         ├─ Anthropic API | AWS Bedrock
                         └─ Jupyter MCP server (same process)

Claude reads/writes/executes cells via MCP tools; the extension provides no custom cell tools.

Troubleshoot

If the frontend extension is not working, check that the server extension is enabled:

jupyter server extension list

If the server extension is installed and enabled but the frontend is not showing up, check that the frontend extension is installed:

jupyter labextension list

Documentation

• docs/USER_GUIDE.md — install & operation guide (start here)
• docs/PLAN.md — step-by-step implementation roadmap
• docs/DECISIONS.md — decision log and rationale
• docs/Architecture.md — component diagram and data flow
• CHANGELOG.md — human-readable milestone log