The 3 MUST Have MCP Servers for Any AI Coding (and How to Use Them)

Key Concepts

• MCP Servers: Modular Component Platforms, used to extend the capabilities of AI coding assistants.
• RAG (Retrieval-Augmented Generation): A technique to improve the quality of AI-generated text by retrieving relevant information from a knowledge base.
• AI Coding Assistants: Tools like Windsurf, Cursor, and Rue Code that help developers write code more efficiently using AI.
• Superbase: An open-source Firebase alternative that provides a backend-as-a-service platform.
• Pydantic AI: A library for building AI agents with data validation and settings management.
• Brave Search API: A web search API optimized for AI applications.
• Streamlit: An open-source Python library for creating interactive web applications.
• Knowledge Base: A repository of information used by AI models to generate responses.
• Embeddings: Numerical representations of text used for semantic search and similarity analysis.

AI Coding Workflow with MCP Servers

The video outlines a workflow for using AI coding assistants, enhanced by Modular Component Platform (MCP) servers, to accelerate software development. The core idea is to augment the AI's capabilities with external knowledge, database management tools, and web search functionalities.

1. Bringing External Knowledge with RAG MCP Server

• Problem: AI coding assistants often lack up-to-date knowledge of specific libraries, tools, or documentation. Built-in documentation support in some IDEs is insufficient.
• Solution: Use an MCP server to provide a Retrieval-Augmented Generation (RAG) knowledge base.
• Recommended Tool: The presenter's own open-source "crawl for AI" RAG MCP server.
  • Functionality: Crawls websites and documentation, storing the information in a private Superbase database.
  • Benefits: Creates a private, manageable knowledge base tailored to specific needs.
  • Setup: Detailed instructions are provided in the server's README file, covering installation, database configuration, and website crawling.
• Alternative: Contact 7, an out-of-the-box solution with thousands of pre-ingested libraries.

2. Database Management with Natural Language

• Problem: AI coding assistants need to be able to create and manage databases as part of the application development process.
• Solution: Use an MCP server to manage the database with natural language.
• Recommended Tool: Superbase MCP server.
  • Functionality: Creates tables, lists projects, and writes SQL queries using natural language.
  • Benefits: Simplifies database management, allowing the AI to build database structures as part of the application.
  • Example: The demo shows the AI creating a "rag_pages" table in Superbase with columns for ID, URL, chunk number, and embeddings.
• Alternative: Neon, a serverless Postgres platform with a similar MCP server.
• Future Trend: Expect more databases to offer MCP servers for natural language management.

3. Web Search for Supplemental Resources

• Problem: AI coding assistants may need to access information beyond the knowledge base, such as community forum posts or documentation not included in the RAG.
• Solution: Use an MCP server to search the web.
• Recommended Tool: Brave MCP server.
  • Functionality: Provides AI-centric web search, summarizing information for LLMs.
  • Benefits: Supplements the knowledge base with up-to-date information and community insights.
  • Usage: Used in tandem with the RAG MCP server to find documentation and examples.
• Brave API: The Brave API is affordable and powerful, offering a generous free tier.

4. MCP Server Configuration in AI IDEs

• Process:
  1. Install and run each MCP server locally.
  2. Configure the AI IDE (e.g., Windsurf, Cursor) to connect to the servers.
  3. Add each server's endpoint to the IDE's MCP configuration file (usually a JSON file).
  4. Provide necessary API keys or access tokens for each server.
  5. Refresh the MCP configuration in the IDE to load the new tools.
• Example (Windsurf):
  • Access the MCP configuration by clicking the hammer icon in cascade mode.
  • Add the crawl for AI server endpoint (e.g., http://localhost:8000).
  • Configure the Brave search server with the Brave API key.
  • Configure the Superbase server with the access token.

5. Live Build Demo: RAG AI Agent with Pydantic AI and Superbase

• Goal: Build a simple RAG AI agent that ingests local files into a Superbase knowledge base and provides a Streamlit interface for querying the agent.
• Setup:
  • Windsurf rules are set up to guide the AI assistant.
  • A planning file outlines the project components.
  • A task file lists the individual tasks to be completed.
• Prompting the AI:
  • Provide a clear description of the desired AI agent.
  • Include examples of Streamlit interfaces and SQL queries for reference.
  • Instruct the AI to use the Superbase MCP server to create database tables.
  • Instruct the AI to use the crawl for AI server to get documentation for Pydantic AI and Superbase.
  • Instruct the AI to use the Brave MCP server for supplemental web search.
  • Explicitly tell the AI to use the MCP servers at the start of the process.
• Build Process:
  1. The AI assistant reads the planning and task files.
  2. The AI assistant uses the crawl for AI server to retrieve documentation.
  3. The AI assistant uses the Brave MCP server to find examples.
  4. The AI assistant creates the necessary files, including the Pydantic AI agent, Streamlit interface, and database setup scripts.
• Iteration and Refinement:
  • The AI assistant may not always complete all tasks correctly on the first attempt.
  • The presenter had to re-prompt the AI to create the database tables using the Superbase MCP server.
  • The presenter had to fix some issues with the document embedding process.
• Final Result:
  • A working RAG AI agent with a Streamlit interface.
  • The agent can ingest text and PDF files into a Superbase knowledge base.
  • The agent can answer questions based on the knowledge base.

6. DataButton: AI App Builder for Businesses

• Overview: DataButton is an AI app builder that handles the entire development lifecycle, from project initiation to deployment.
• Functionality:
  • Generates a full plan with tasks based on project requirements.
  • Integrates with authentication, databases, payments, and storage services.
  • Takes care of both front-end and back-end development.
  • Deploys agents and APIs behind the scenes.
• Benefits:
  • Reduces the need for a CTO.
  • Enables lean businesses to compete with larger companies.

7. Notable Quotes

• "You can transform any AI coding workflow with these servers no matter what you want to build."
• "We're giving superpowers to our AI coders."
• "Data button really is a game changer for lean businesses that are looking to leverage AI to compete with companies that are 10 times their size."

8. Conclusion

The video demonstrates how MCP servers can significantly enhance AI coding workflows by providing external knowledge, database management tools, and web search capabilities. By leveraging these tools, developers can accelerate the development process and build more sophisticated applications with AI assistance. The live build demo showcases the practical application of these concepts, resulting in a functional RAG AI agent built in a short amount of time. The presenter also highlights DataButton as a tool that further streamlines AI app development for businesses.