Model Context Protocol (MCP)

Retrieval and context get the right information into a model. Tools let a model do things — query a database, send an email, open a file. The problem was that every app wired up tools its own way. The Model Context Protocol (MCP), open-sourced by Anthropic in late 2024, standardises that wiring.

MCP is often called the "USB-C for AI": an open standard where a host application (the MCP client) connects to any number of MCP servers over JSON-RPC. Each server exposes three kinds of primitive — Tools (actions the model can call), Resources (data it can read), and Prompts (reusable templates). Write a server once, and every MCP-aware application can use it.

Example. Instead of building a bespoke GitHub integration inside your chat app, your IDE, and your automation, you run one GitHub MCP server. Claude Code, your editor, and a workflow tool all speak to it through the same protocol — and the work transfers. Personal runtimes lean on the same idea of a shared tool surface: OpenClaw exposes 50+ integrations through its gateway, and Hermes Agent gives every subagent its own toolset, so capability is added once and reused everywhere.

The governance point is sharp: a protocol that lets a model act in the world makes least-privilege, scoping, and approval design non-negotiable. The convenience of MCP and the security demands of agents arrive on the same day.

References & further reading

• Anthropic (2024). Introducing the Model Context Protocol. Anthropic News.
• MCP Project (2025). Model Context Protocol — Specification. modelcontextprotocol.io.
• Anthropic (2025). Writing Effective Tools for AI Agents. Anthropic Engineering.
• Anthropic (2024). Introducing Computer Use. Anthropic News.