Workshop: 2 days
Building MCP Servers
Workshop description
The Model Context Protocol (MCP) is revolutionizing how AI systems interact with external data and services. This hands-on workshop teaches you to build sophisticated MCP servers that can provide AI models with custom tools, dynamic resources, and contextual information. From basic server setup to advanced features like authorization and deployment, you'll gain the skills to create MCP servers that unlock new possibilities for AI-powered applications.
What is MCP?
We'll explore what the Model Context Protocol is and the main applications for extending AI capabilities.
MCP Lifecycle
Understanding the complete lifecycle of an MCP session from initialization to shutdown, including connection management and error handling.
Transports
STDIO vs HTTP communication methods: how to actually communicate via MCP, exploring the advantages of each transport mechanism.
Tools
Build the most powerful feature of MCP: tools that allow LLMs to autonomously request additional context and perform actions on behalf of users.
Resources
Learn how users can inject context from your MCP server before sending prompts. We'll explore different resource types and create dynamic resources.
Prompts
Help users with pre-prepared prompts they can include with a couple of clicks. Learn how to expose reusable prompts from your MCP server.
Completions
Implement completion functionality for both Resources and Prompts to provide MCP clients with information about what's available on your server.
Elicitation
Handle scenarios where MCP servers need extra user information (like GitHub usernames) by implementing elicitation to ask users directly.
Sampling
Learn when and how to use sampling to ask users for permission to use their LLM to fulfill requests, managing costs and permissions effectively.
Authorization
Secure your MCP server with proper authorization so users can only access their own data, plus deployment strategies for sharing via npm or hosting.
Deployment
MCP servers are stateful by nature so deploying them to serverless needs a bit more carefulness...let's learn the ins and outs of how to deploy an MCP server.
Your mentor
