What is MCP? — Model Context Protocol in 12 Minutes

Key Takeaways

• MCP defines a versioned, JSON-schema-validated contract between an LLM host and external tools — think of it as the service-mesh sidecar for AI tool invocation: discovery, auth context, and structured I/O in one protocol.
• Tool registration in MCP is declarative: each tool publishes an input/output JSON schema to the MCP server, enabling the LLM to perform capability discovery at runtime without hard-coded API knowledge.
• Context injection is MCP's primary production value — it lets the host process selectively push resources, memory state, and retrieved documents into the LLM context window through a single, auditable channel rather than ad-hoc prompt concatenation.
• Memory anchor for exam pressure: MCP is to LLM-tool integration what an API gateway is to microservices — one versioned, schema-enforced entry point that handles routing, capability negotiation, and context relay.