Model Context Protocol
MCP
Open standard from Anthropic (Nov 2024) for connecting AI assistants to external tools, data sources, and prompts. Think "USB for LLM tools" — one protocol, many servers, any client.
MCP is a JSON-RPC-style protocol that standardises how an LLM client (Claude Desktop, Cursor, Continue, etc.) connects to tool servers. Before MCP, every IDE/agent had its own plugin format. After MCP, you write one server (e.g. a database query tool) and it works in every MCP-aware client.
Servers expose three primitives: tools (functions the model can call), resources (read-only data the model can attach), and prompts (template snippets the user can summon). Transport is stdio or HTTP+SSE.
Adoption accelerated through 2025-2026 as the major IDEs and agent frameworks all landed support. Most production agents now use MCP for tool-use rather than custom integrations.
See also
-
Agentic
Agentic systemsLLM-driven loops that plan, take actions in the world (call tools, edit files, hit APIs), observe results, and iterate — rather than just answering a single prompt. The dominant 2026 paradigm for AI engineering.
-
Tool use
Tool use / function callingThe LLM emits a structured request to call an external function (search, calculator, API), the host runs it, the result goes back in the next turn. Foundation of every agent worth shipping.