Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, that defines how AI applications (like Claude Desktop, Claude Code, or any MCP-compatible client) connect to external tools, data sources, and services in a consistent, reusable way. Before MCP, every AI product that wanted to read your Google Drive, query your database, or control your browser had to write a bespoke integration. MCP standardizes this into a client-server protocol — similar in spirit to how the Language Server Protocol standardized editor-to-language-tooling communication — so a single MCP server (say, a Postgres server or a GitHub server) can be plugged into any MCP-compatible AI client without custom glue code. Why it matters for AI/SaaS builders: it's rapidly becoming the default way to give an LLM agent "hands" — safe, permissioned access to real systems — without hand-rolling function-calling schemas for every tool from scratch. If you're building an AI feature into a SaaS product, exposing your product's capabilities as an MCP server means any MCP-aware AI client (not just your own chat UI) can use it, which is a meaningful distribution and integration advantage. How it works: an MCP server exposes three primitive types — Tools (functions the model can call, like `create_ticket` or `run_query`), Resources (data the model can read, like a file or a database row), and Prompts (reusable prompt templates). The MCP client (the AI application) discovers what a server offers via a handshake, then the model decides at runtime which tools to invoke based on the user's request, with the client mediating permissions and execution. Transport is typically JSON-RPC over stdio (for local servers) or HTTP/SSE (for remote servers). Worked example: a developer connects Claude Code to a Postgres MCP server for their SaaS app's database. They ask, "Which users signed up this week but never verified their email?" Claude, via MCP, calls the server's `query` tool with a SQL statement it composes (`SELECT id, email FROM users WHERE created_at > now() - interval '7 days' AND email_verified_at IS NULL`), the MCP server executes it against the real database and returns rows as a Resource, and Claude summarizes: "14 users signed up this week without verifying — here's the list, and I can draft a reminder email if you'd like." No custom API endpoint was built for this — the MCP Postgres server made the database generically queryable by any compliant AI client.