Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open standard introduced by Anthropic in late 2024 that defines how AI assistants connect to external data sources and tools via a standardised server interface.
Model Context Protocol (MCP) is an open standard introduced by Anthropic in late 2024 that defines how AI assistants connect to external data sources and tools via a standardised server interface. Adoption has been broad: Claude Desktop, ChatGPT (via plugin runtime), Cursor, Continue, Zed, and dozens of agent frameworks now consume MCP servers.
The mechanic: a website or service exposes an MCP server that advertises capabilities (read documents, search a knowledge base, execute a query). AI assistants discover and call those capabilities at runtime, often by reading a small `mcp.json` manifest from a known path. Where llms.txt is a one-way "here's our content" hint, MCP is a two-way "here's how to query our content" contract.
For AEO purposes, MCP is the agent-native equivalent of the public website. A site with a good MCP server is directly callable by an AI agent — the agent can answer "what does Acme charge for the enterprise tier" without having to crawl the pricing page and parse it. As agents proliferate, brands with MCP servers will be cited via agent calls, not just citation links.
AEOlens does not yet score MCP server presence in the structural audit — it is a forward-looking signal that doesn't materially affect citation rate today (June 2026). Expect that to change in the next 12–24 months as agent traffic grows.
Frequently asked
Does my website need an MCP server today?
For most marketing sites, no — the agent traffic isn't material yet. The exception is developer-facing products: dev tools, APIs, infrastructure platforms. Cursor, Continue, and Claude users routinely call MCP servers for those categories, and a missing server means a missing citation.
How is MCP different from a REST API?
A REST API is consumed by your code; MCP is consumed by an AI agent at runtime. MCP servers advertise capabilities in a standard schema the agent can read without you publishing documentation, and they expose tool calls (not just data endpoints) so the agent can act on behalf of the user.
See how AI engines read your own site
Run a free AEO audit on any URL — 48 structural checks, score in under 60 seconds, no signup.
Run a free audit