Model Context Protocol (MCP)
An open protocol that connects AI applications to external tools and data through one standard interface — so integrations are written once and used by any capable client.
01What it means
The Model Context Protocol (MCP) is an open standard for connecting AI applications to external systems — tools, data sources, and services — through a single, uniform interface. Instead of building a bespoke integration for every combination of model and tool, you run an MCP server that advertises its capabilities, and any MCP-capable client (an IDE, an assistant, an agent harness) can discover and call them. The protocol keeps a hard line on isolation: a server sees requests, never the conversation around them, the other servers in use, or the host application.
Mechanically, a server declares tools with typed schemas; the model decides when to call which tool and with what values; the server executes and returns the result. Transport follows a simple rule: a command means a local server over stdio, a URL means a remote one over streamable HTTP. This site runs its own read-only example — nine tools exposing our public content at /api/mcp, with a server card at /.well-known/mcp/server-card.json.
For buyers, MCP turns "does it integrate with our AI stack?" from a per-vendor question into a standards question. A vendor with an MCP surface has done the integration work once, for every compliant client at once.
02Why it matters for buyers
- —Integration stops being bespoke: any MCP-capable client in your stack can use a compliant tool — the integration your vendor built once is the one you consume everywhere.
- —Security review has a perimeter: isolation is built into the protocol (a server sees requests, not conversations), so scoping an audit is tractable.
- —Ask vendors where their MCP surface lives: publishing one is a concrete signal they take agent access seriously — it is one of the checks in our agent-readiness tool.
03Common questions
What does MCP stand for, and who is behind it?
Model Context Protocol — an open protocol introduced by Anthropic in late 2024 and since adopted across model vendors, IDEs, and agent frameworks. It is a specification anyone can implement, on either side.
Is MCP only for Claude?
No. It is an open protocol: any client that speaks it — IDEs, agent harnesses, other vendors' assistants — can connect to any compliant server. That vendor-neutrality is most of its value.
What is the difference between MCP and an API?
An API is a product-specific contract; MCP is a standard envelope for discovering and calling capabilities — names, typed schemas, and results — so clients can use a server they have never seen before. Most MCP servers wrap existing APIs rather than replace them.