01. What is MCP?

Chapter 1 of 22 · 15 min

KEY INSIGHT

MCP standardizes how AI systems interact with external tools, resources, and prompts through a defined protocol rather than custom integrations. Model Context Protocol (MCP) is an open specification that defines how AI applications communicate with external capabilities. Instead of building custom integrations for every tool an AI might need, MCP provides a standardized interface that works across different AI frameworks and implementations. The core problem MCP solves is integration fragmentation. Without a protocol, connecting an AI assistant to a file system, database, or web API requires unique code for each connection. If the AI application changes, all those custom integrations break. MCP abstracts this by defining message formats, transport mechanisms, and capability negotiation that remain stable across implementations. MCP has three primary capability types: **Resources** expose data to AI applications. A resource might be a file, database query result, or API response. AI systems request resources when they need contextual information for their responses. **Tools** enable AI systems to perform actions. Unlike resources, tools modify stateΓÇöcreating files, executing queries, posting messages. AI applications call tools based on user requests and can describe results back to the user. **Prompts** provide reusable prompt templates. These pre-configured prompts help standardize complex workflows, ensuring AI applications respond consistently for common tasks. The protocol operates over HTTP or stdio, with JSON-RPC 2.0 as the message format. Clients discover server capabilities during connection initialization, then use those capabilities through defined method calls. MCP servers are language-agnostic. The official SDK exists for TypeScript and Python, but any implementation that follows the protocol specification works with any MCP-compatible client. Understanding MCP matters because it shifts integration work from custom code to configuration and standards. Building one MCP server provides connectivity to any AI client that speaks the protocolΓÇöfuture-proofing integrations against framework changes.

EXERCISE

List three external services (APIs, databases, file systems) that an AI assistant might need to access. For each, describe what capability type (resource, tool, or prompt) would be most appropriate and why.