
jangtrinh/colab-mcp-recipes
Developer toolsThe definitive collection of AI agent workflows for Google Colab MCP. 106 ready-to-use recipes across GPU/Deep Learning, Generative AI, Data Science, Google Integration, Education & more.
Overview
Pick any recipe below, paste the prompt into your AI agent, and watch it build a full notebook on Google's cloud. These 5 recipes showcase the best of Colab MCP: Gemini CLI · Claude Code · Cursor · VS Code Copilot · Any MCP-compatible client 💡 Colab MCP = ssh for the AI age. Your agent writes code, runs it on cloud GPUs, debugs it, and reports back — in one conversation. Read the full deep-dive → Click a category to expand. Each recipe includes a copy-paste prompt template. 🆓 Free GPU/TPU — Deep Learning & Training (13) 🆓 Free GPU/TPU — Generative AI & LLMs (14) 📦 Pre-installed ML Ecosystem — Data Science (16)
README
Curated by Jang Trịnh
⚡ Quick Start
3 steps. Under 2 minutes.
1. Install the MCP server
pip install uv
2. Add to your agent’s MCP config
{
"mcpServers": {
"colab-mcp": {
"command": "uvx",
"args": ["git+https://github.com/googlecolab/colab-mcp"],
"timeout": 30000
}
}
}
📖 Setup guides: Gemini CLI · Claude Code · Cursor / VS Code
3. Copy a recipe prompt and go!
Pick any recipe below, paste the prompt into your AI agent, and watch it build a full notebook on Google’s cloud.
⭐ Start Here — Top 5 Picks
Not sure where to begin? These 5 recipes showcase the best of Colab MCP:
| Recipe | Why this one | |
|---|---|---|
| 🔥 | CSV → Insights | Works instantly — upload any CSV, get a full analysis notebook |
| 🧠 | Run Llama 3 | Experience free GPU power: run a real LLM with one prompt |
| 🎨 | Stable Diffusion | Visual wow: generate images from text on Colab’s GPU |
| 📊 | A/B Test Analysis | Business-ready: statistical significance, power analysis, Bayesian |
| 🔗 | Colab → GitHub | Multi-MCP magic: train model → auto-commit results to repo |
🏗️ How It Works
graph LR
A[You] -->|Prompt| B[AI Agent]
B -->|MCP Protocol| C[Colab MCP Server]
C -->|Creates & Runs| D[Google Colab Notebook]
D -->|Free GPU/TPU| E[Results & Visualizations]
style A fill:#4285F4,stroke:#333,color:#fff
style B fill:#EA4335,stroke:#333,color:#fff
style C fill:#FBBC04,stroke:#333,color:#000
style D fill:#34A853,stroke:#333,color:#fff
style E fill:#4285F4,stroke:#333,color:#fff
Works with: Gemini CLI · Claude Code · Cursor · VS Code Copilot · Any MCP-compatible client
🔥 Why Colab MCP?
| Without MCP | With Colab MCP |
|---|---|
| Copy-paste code between AI chat and Colab | Agent writes AND runs code directly |
| “Run this locally” = no GPU, OOM errors | Free T4 GPU, 12+ GB RAM, sandboxed |
| Results stuck in chat = lost | Persists in a shareable .ipynb notebook |
| One tool at a time | Chain Colab + GitHub + Sheets + Slack |
💡 For hackers: Colab MCP =
sshfor the AI age. Your agent writes code, runs it on cloud GPUs, debugs it, and reports back — in one conversation. Read the full deep-dive →
📚 Full Recipe Catalog (106)
Click a category to expand. Each recipe includes a copy-paste prompt template.
🧠 System Prompts
Supercharge your recipes with pre-built agent personas:
| Prompt | Best For | What It Does |
|---|---|---|
| Data Analyst | Data Science recipes | Thinks in pandas, asks clarifying questions about data |
| ML Engineer | GPU/Training recipes | Optimizes for GPU memory, picks best architectures |
| Researcher | Research recipes | Structures experiments, cites papers, tracks results |
| Educator | Education recipes | Explains concepts, adds interactive widgets, progressive difficulty |
💡 Pro tip: Combine a system prompt + a recipe for best results. Example: Use the ML Engineer prompt with the Quick Fine-tune recipe.
🤝 Contributing
We welcome contributions! See CONTRIBUTING.md for:
- 📝 Recipe template and format guidelines
- ✅ Quality checklist for new recipes
- 🏷️ How to add new categories
Every recipe should be something you’d star. ⭐
📜 License
PolyForm Shield 1.0.0 © Jang Trịnh — free to use and modify; no resale or competing products.
If this helped you, give it a ⭐ — it helps others discover these recipes!
LinkedIn · Official Colab MCP · MCP Protocol
Made with 🧡 by Jang Trịnh — Founder @ EaseUI · Ex-Head of Design @ Elfie
Install
uvx git+https://github.com/googlecolab/colab-mcpConfiguration
{
"mcpServers": {
"colab-mcp": {
"command": "uvx",
"args": ["git+https://github.com/googlecolab/colab-mcp"],
"timeout": 30000
}
}
}