Plug Claude into the rest of your stack
MCP (Model Context Protocol) is the open standard for giving an agent capabilities beyond its built-in tools. An MCP server exposes a function set; Claude calls those functions like any tool. The result: Claude can query Sentry for live errors, post Slack messages, read Notion pages, run a Postgres query, or talk to any of 1,800+ servers in the public ecosystem.
Transports: stdio spawns a local process (best for tools needing your filesystem), HTTP / Streamable HTTP talks to a remote URL (the recommended path for SaaS — 2025-06-18 spec), SSE is the legacy remote transport (deprecated, but you'll see it in older configs).
Scopes: local (default, this project, stored in ~/.claude.json), project (committed to .mcp.json, shared with the team), user (across all your projects). Pick scope by who else needs the connector.