MCP (Model Context Protocol) is an open standard, introduced by Anthropic, for connecting AI assistants to external systems. In Claude Code, you add MCP servers that expose tools (actions the model can call), resources (data it can read), and prompts (reusable templates) — letting the agent talk to databases, issue trackers, browsers, design tools, and internal APIs without bespoke wiring for each one.
An MCP server is a separate process Claude Code communicates with over a defined transport — commonly a local stdio subprocess, or a remote HTTP/SSE endpoint. Once a server is configured, its tools appear alongside Claude Code's built-in capabilities, and you govern whether the agent can invoke them via permissions. Servers can be scoped to just you, shared with a project, or made available across all your projects, making them a core way Claude Code is extended beyond the local filesystem and shell.
Because MCP is an open protocol rather than a Claude-only feature, the same server can be reused across other MCP-compatible clients and editors. For the exact commands and config keys to add, list, and manage servers, follow the official Claude Code docs — they stay current as the CLI evolves. For a deeper walkthrough, see our MCP explainer.
Why it matters
MCP is how Claude Code reaches the rest of your stack — pulling live context from your database, ticketing system, or docs and taking real actions through one standard interface, so the agent works against your actual systems instead of just the files in your repo.
Deep dive: Claude Code MCP Explained: The Model Context Protocol Guide
Related terms
See MCP (Model Context Protocol) in action across releases — browse the Claude Code changelog.
Get release alerts