Author, edit, and execute Google Colab notebook cells through a browser MCP bridge.
npx skills add https://github.com/bfairkun/dotfiles --skill colab-mcp-bfairkun
Please help me install this Agent Skill. Skill: colab-mcp Source: https://github.com/bfairkun/dotfiles/tree/main/agents/.agents/skills/colab-mcp Command: npx skills add https://github.com/bfairkun/dotfiles --skill colab-mcp-bfairkun
add_code_cell, update_cell, and run_code_cell. - Shared Kernel Execution: Run Python, R, or Julia cells against a persistent kernel with pre-installed data science packages (pandas, scikit-learn, torch, jax, and more). - Drive Notebook Support: Follow in-tab navigation to open and edit existing notebooks stored in Google Drive after connecting. - Use Case: Ask the agent to build an exploratory data analysis in Colab; it opens a notebook in your Chrome tab, writes cells, runs them, and you watch plots and tables render natively. ## Quick Start Connect to a live Colab notebook in my browser and create a notebook that loads a CSV and plots its distributions.High-intent search queries and answers about installing and using this skill.
Call the open_colab_browser_connection MCP tool, which opens a Colab tab in Chrome with a proxy token. The user must complete the handshake within 60 seconds, with Chrome already running and signed into Google, or the bridge breaks for the session.
The bridge exposes add_code_cell, add_text_cell, update_cell, delete_cell, move_cell, get_cells, and run_code_cell. These appear dynamically after the WebSocket connection is established with the Colab frontend.
Safari silently blocks the localhost WebSocket from colab.research.google.com under Private Network Access rules. Use a Chromium browser like Chrome, Brave, or Arc, or set the BROWSER environment variable on the MCP entry to spawn Chrome directly.
Yes. The entry tool always opens a scratch notebook, but the bridge follows in-tab navigation. Connect first, then use File → Open notebook → Drive in the Colab UI, and subsequent MCP calls operate on that notebook.
run_code_cell blocks until completion, so long-running cells tie up the bridge. There is no state-introspection tool, the free tier has 2 CPUs and no GPU, and changing runtime type requires manual action in the Colab UI.
shammunDefines a four-tier data sourcing policy for fluid dynamics benchmarks, datasets, and reference values.
zongtingweiPackage analysis outputs into reproducible reports and publication-ready figure exports.
edwinhuInspect .ipynb files for tracebacks and cell execution issues.
agentic-inRun interactive Jupyter notebook sessions with live kernel management.