The tool-integration layer

MCP is an open standard for connecting AI agents to tools and data. A server describes what a system can do, and any MCP-speaking agent can discover and call it without a custom integration. Anthropic released it in November 2024 and the major model providers have since adopted it. (Model Context Protocol)

The server publishes a list of tools, each with a name, a description and a JSON schema for its inputs. The agent asks for that list, decides which tool to call, and sends the arguments back. The server runs the action and returns the result.

The protocol does not make a tool safe by itself. A server runs with whatever access you grant it, and content it returns can carry prompt injection aimed at the agent. Run only servers you trust, scope credentials narrowly, and require approval before anything irreversible. (OWASP)

A REST API is built for developers to read docs and write code against. MCP describes the same actions in a machine-readable way, so an agent can find and call them at runtime with no integration written first. Plugins were tied to one host app; MCP works with any compatible agent.

Start with the vendor docs for the software you already use, since most major vendors now ship a first-party server. Beyond that there are moderated catalogues from Docker and GitHub, plus the official MCP registry. Open self-publish directories list far more but verify nothing. (MCP registry)