Overview
- Model Context Protocol (MCP) is an open standard from Anthropic that defines a single, consistent way for large language models to discover and call external tools and data sources.
- MCP uses a three-part architecture — a Host (the app), a Client (the connector inside the host), and a Server (the tool provider) — so hosts only need to speak one protocol to access many tools.
- A standard MCP interaction has the host list available tools to the LLM, the LLM selects a tool, the host calls that tool through the MCP server, and the tool’s result is returned to the LLM as grounded context for its final reply.
- Coverage on Monday reported a forthcoming protocol tweak that makes session handling more stateless so MCP servers can scale behind load balancers without each machine tracking per-session IDs, which should lower operational complexity for providers.
- Writers compare MCP to REST or USB-C for AI because it moves integration work to tool-side MCP servers, cutting host maintenance and letting developers build agentic assistants that can safely read files, query databases, update services, and trigger workflows.