Definition
MCP (Model Context Protocol)
MCP, the Model Context Protocol, is an open standard created by Anthropic for connecting AI assistants to external tools and data sources. MCP servers expose capabilities — database queries, API calls, browser automation, file operations — that any MCP-compatible agent can use. Cursor, Claude Code and Codex all support it, so one server serves all three.
Why it matters
MCP is what stops tool integrations being written three times. A server built once — for your database, your issue tracker, your internal API — works in every compatible agent, so switching tools does not mean rebuilding the connections around them.
How each tool handles it
- Cursor
- Configured in settings.
- Claude Code
- Configured through
.mcp.jsonorclaude mcp add. - Codex
- Configured in
~/.codex/config.toml, shared across its surfaces.
Frequently asked questions
What is an MCP server?
A program that implements the MCP specification and exposes one or more tools to AI agents. Examples include GitHub MCP for pull requests and issues, Postgres MCP for database queries, Puppeteer MCP for browser automation and Context7 MCP for library documentation.
Do all three tools support MCP?
Yes. Cursor, Claude Code and Codex all support MCP servers; only the configuration location differs. That is the point of a protocol — the server does not need to know which agent is calling it.
How is MCP different from Agent Skills?
Agent Skills are lighter. A skill augments a single capability; an MCP server maintains a persistent connection to an external system. Skills install with npx skills add <owner/repo>; MCP servers are configured per tool and run as processes.