In this hands-on workshop, we’ll go beyond theory and build a real AI agent setup using MCP – Model Context Protocol – to create an extensible, plug-and-play ecosystem between LLMs and APIs.
We’ll start with the basics: setting up Claude Desktop or Cursor IDE, wiring up an MCP server (Qdrant) to persist generated code, and then extending the system with a second tool (e.g., Airbnb search or Google Maps). Along the way, we'll cover:
* How MCP clients and servers interact
* STDIO and Server-Sent Events transport layers
* Writing and running multi-server configs in JSON
* Saving & retrieving LLM responses via Qdrant vector DB (RAG)
* Building custom tools to expose your own APIs to the LLM
This is "vibe coding" in action – collaborative, contextual, intelligent development where your model remembers, learns, and integrates seamlessly.
Workshop Agenda:
1. Introduction to MCP Servers (15 minutes)
* Overview of MCP: What is Model Context Protocol (MCP)? Why is it important for modern development workflows?
* Key Components of MCP:
* Client, Server, Transport Layer, and Protocol.
* How these components work together for enhanced model context and interaction.
* Real-world Use Cases: Discuss how MCP can help in common scenarios like task automation, integrating databases, and fetching external data.
2. Setting Up and Configuring MCP Servers (25 minutes)
* Demo of MCP Server Setup:
* Walk through how to set up an MCP server with an example like Qdrant (a vector database).
* Hands-on Activity: Attendees will follow along to configure their own MCP server using a JSON configuration file.
* Integrating External Services: How to connect MCP with Google Drive, Dropbox, or other cloud-based services to fetch/store data.
* Structuring JSON Files: Best practices for writing a well-organized MCP configuration that supports multiple servers.
3. Building a Real-World Use Case: Automating Tasks with MCP (30 minutes)
* Example Use Case: Build an agent using MCP to automate tasks and store/retrieve data using a vector database like Qdrant.
* Hands-on Activity:
* Attendees will use MCP to integrate services like Airbnb and Qdrant, automating repetitive tasks and generating context-aware responses.
* Example tasks: Automating fetching and storing data related to vacation rentals and previous code templates.
* Vibe Coding in Action: How MCP servers can enhance Vibe coding by enabling seamless task execution and context retrieval for future coding sessions.
4. Best Practices for Optimizing MCP Workflow (20 minutes)
* Optimizing Transport Layers: Discuss tips on enhancing the speed and reliability of communication between servers and clients.
* Troubleshooting Common Issues: How to handle errors, debug MCP server setups, and optimize your workflow.
* Structure and Efficiency: How to organize your JSON files and handle multiple servers in a single file for better scalability.
5. Q&A and Open Discussion (20 minutes)
* Open the floor for attendees to ask questions, provide feedback, and share experiences.

By the end of this workshop, participants will:
1. Understand the fundamental concepts of MCP and how it works.
2. Gain hands-on experience with setting up and configuring MCP servers.
3. Learn how to integrate vector databases and external APIs into their workflow using MCP.
4. Automate tasks and retrieve meaningful context for more efficient coding using MCP.
5. Implement best practices for managing and optimizing MCP workflows.
* A fully working MCP setup with Qdrant + Claude
* A personalized AI coding assistant that remembers your code patterns
* Enough knowledge to build and expose your own APIs via MCP