MCP = Next Big Opportunity? EASIST way to build your own MCP business

Key Concepts

• mCP (Machine Code Protocol): A unified way for AI agents or AI applications to access external assistance.
• AI Agents: Software programs designed to perform tasks autonomously, often leveraging large language models (LLMs).
• Function Calling: A mechanism by which LLMs can trigger external functions or APIs to perform actions in the real world.
• LLMs (Large Language Models): AI models trained on vast amounts of text data, capable of generating human-like text, translating languages, and answering questions.
• Go-to-Market Strategy: A plan that outlines how a company will reach its target customers and achieve a competitive advantage.

mCP: A Unified Protocol for AI Agents

The video discusses the emergence of mCP as a standardized protocol for AI agents to interact with external systems. It draws a parallel to the early days of the internet and the adoption of TCP/IP, which unified communication between devices.

• Problem: LLMs from different providers (OpenAI, Claude, Gemini, Llama) have different formats for function calling, making integration complex.
• Solution: mCP provides a unified format for communication, simplifying the development of AI agent clients.
• Analogy: mCP is to AI agents what TCP/IP was to the internet.
• Impact: Lower barrier to entry for building AI agent clients (like Cursor, Wingman), fostering innovation.

Startup Opportunities in the mCP Ecosystem

The video highlights several startup opportunities arising from the growth of the mCP ecosystem:

• AI Agent Clients: Building AI agents for specific verticals (sales, customer support) or general purposes.
• mCP Marketplaces: Platforms for discovering, testing, and curating mCPs (e.g., GLlama, Smithy).
• mCP Development & Monetization: Creating and distributing mCPs to enhance existing AI agent clients, with monetization strategies.
• Example: 21st.co building mCPs to enhance coding experience in Cursor and Wingman, and monetizing them.

Building an mCP from Scratch: A Step-by-Step Guide

The video provides a practical guide to building an mCP using the Python SDK.

1. Install the mCP SDK: pip install mCP
2. Create a new Python file: e.g., test_mcp.py
3. Import the FastMCP class:

   from mcp import FastMCP
   

4. Initialize the mCP server:

   mcp = FastMCP()
   

5. Define a tool using the @mcp.tool decorator:

   @mcp.tool
   def calculate_bmi(weight_kg: float, height_cm: float) -> float:
       """
       Calculates BMI given weight in kilograms and height in centimeters.
   
       Args:
           weight_kg (float): Weight in kilograms.
           height_cm (float): Height in centimeters.
   
       Returns:
           float: The calculated BMI.
       """
       height_m = height_cm / 100
       bmi = weight_kg / (height_m ** 2)
       return bmi
   

6. Run the mCP server: mcp run
7. Configure the mCP in your AI agent client (e.g., Cursor):
   • Add a new mCP.
   • Set the command to mcp run.
   • Paste the path to your Python file.

Case Study: Building a Figma mCP

The video demonstrates building a Figma mCP to extract design information.

1. Figma API: Utilizes the Figma API, specifically the "Get File Nodes" endpoint.
2. Authentication: Requires a Figma API token (generated in Figma settings).
3. Helper Function (fetch_figma_nodes): Makes API calls to Figma.
4. mCP Tool (get_node): Exposes the Figma API functionality to AI agents.

   @mcp.tool
   def get_node(file_key: str, node_id: str) -> dict:
       """
       Retrieves a specific node from a Figma file.
   
       Args:
           file_key (str): The key of the Figma file.
           node_id (str): The ID of the node to retrieve.
   
       Returns:
           dict: The JSON representation of the Figma node.
       """
       return fetch_figma_nodes(file_key, node_id)
   

5. Data Cleaning: Implements data cleaning to provide more relevant information to the LLM (e.g., converting colors to simpler formats).
6. Example Prompt: "Implement UI like figma mop here in Pixel Perfect manner in index.html"

Distributing Your mCP

The video outlines how to distribute your mCP to others.

1. Documentation: Create a requirements.txt file listing dependencies and a README.md file explaining how to run the mCP.
2. GitHub Repository: Upload the project to GitHub.
3. mCP Marketplaces: Submit your mCP to platforms like GLlama or Smithy.

Go-to-Market Strategy for New Products

The video briefly mentions the importance of a solid go-to-market strategy for launching new products like mCPs. It recommends a free go-to-market playbook by Hosital, covering:

• Designing a launch blueprint.
• Identifying market signals.
• Crafting a market story.
• Building a viral tweet.
• Tracking product launch metrics.

Conclusion

mCP represents a significant opportunity for standardizing AI agent interactions with external systems. The video provides a practical guide to building and distributing mCPs, highlighting the potential for new businesses and innovation in the AI ecosystem. The key takeaways are the simplicity of building mCPs with available SDKs, the importance of data cleaning for effective LLM integration, and the potential for monetization through mCP marketplaces.