🚀 Add Prompt Templates for Better Agent Integration

[ChengJiale150]

📋 Issue Description

The Jupyter MCP Server core functionality is already comprehensive and well-implemented, providing robust tools for AI agents to interact with Jupyter notebooks. However, to transform this from a "usable" tool to an "easy-to-use" and "powerful" tool, we need to add one critical missing component: Prompt Templates.

🔍 Current State Analysis

✅ What's Already Excellent

• Complete Tool Set: 15+ specialized tools covering all aspects of notebook management
• Real-time Execution: Live cell execution with multimodal output support
• Multi-notebook Support: Seamless switching between different notebooks

❌ The Missing Piece

While agents can technically use all these tools, they often struggle with:

• Context Understanding: Not knowing the best practices for different workflows
• Domain Expertise: Lack of specialized prompts for specific use cases (beginner vs expert, different domains)

🎯 Proposed Solution

Add a dedicated prompts/ directory containing curated prompt templates specifically designed for different Jupyter MCP usage scenarios.

📂 Suggested Prompt Categories

prompts/
├── data-scientist/
│   ├── README.md
│   ├── AGENT.md
│   └── ... (more files)
├── kaggle-competition/
│   ├── README.md
│   ├── AGENT.md
│   └── ... (more files)
├── learning-assistant/
│   ├── README.md
│   ├── AGENT.md
│   └── ... (more files)
├── README.md
└── ... (more files)

Each prompt template should include:

• README.md: A brief description of the prompt template
• AGENT.md: The actual prompt template

🚀 Implementation Benefits

For End Users

• Faster Onboarding: New users can quickly understand how to use Jupyter MCP effectively
• Better Results: Domain-specific guidance leads to higher quality outcomes

For AI Agents

• Clear Context: Understand the specific domain and requirements
• Quality Standards: Understand what constitutes good vs. poor output

For the Project

• User Adoption: Lower barrier to entry for new users
• Community Growth: More users will share their templates with the community
• Feedback Loop: Templates can be improved based on real usage patterns

💬 Call to Action

This feature request addresses a critical gap between having powerful tools and being able to use them effectively. By providing curated prompt templates, we can significantly improve the user experience and help both beginners and experts get the most out of Jupyter MCP Server.

Next Steps:

1. Discuss the proposed template categories and structure
2. Start with a minimal set of core templates
3. Gather community feedback on most needed scenarios
4. Iterate based on real usage patterns

[eleonorecharles]

Hi @ChengJiale150, thanks for opening this issue! This is a great proposal, and I completely agree that adding structured prompt templates is a critical next step for usability.

I have one thought on the implementation. Instead of creating a new system of Markdown files, I'm wondering if it would be more appropriate to use the existing MCP Prompt feature from the Python SDK? See: https://github.com/modelcontextprotocol/python-sdk?tab=readme-ov-file#prompts

It seems like this feature is designed for this exact purpose. What are your thoughts on leveraging that instead?

[ChengJiale150]

@eleonorecharles Fully agree with you, the prompt section in MCP is design for this. However, in my using many MCP Client, I discover that not all client support all feature of MCP (e.g. Cursor only support @mcp.tool and @mcp.resource). So I decide to store it to prompt/ instead of @mcp.prompt for fully compatibility for all client.

[eleonorecharles]

@ChengJiale150 could we use the prompts/ directory as the source of truth, and then have the server also automatically parse and register those files as native @mcp.prompt ?

[ChengJiale150]

@eleonorecharles After I research MCP Prompt (https://modelcontextprotocol.io/specification/2025-06-18/server/prompts), I know prompt is triggered by user to easily reuse. So I decide to split prompt to two part:

• system prompt : This part decide who agent is and what agent should do, it should manully load at the very first place of agent context (like CLAUDE.md in Claude Code) and should not triggered by user. So the prompt/ directory will focus on system prompt which define agent itself.

example

# Role
You are Jupyter Agent, ...

• user prompt : This is what @mcp.prompt design for! It can easily trigger by user, details as follows in MCP protocol:

example

/data_clean

Now you should execute data_clean task, here are some tip you should known:
...

What do you think? We can open a new issue to decide what prompt template should be added in @mcp.prompt

[eleonorecharles]

Agree, thanks @ChengJiale150 !