MCP server endpoint - the protocol AI agents use to call your tools
Tests whether your site exposes a Model Context Protocol endpoint that AI agents can find and call.
What this signal tests
We check whether your site exposes an MCP (Model Context Protocol) server - the emerging standard that lets AI assistants like Claude, ChatGPT, and Copilot call your APIs as native tools. We probe the conventional locations (/mcp, /sse, /api/mcp, /.well-known/mcp) and look for the Streamable HTTP transport signature.
Why it matters for your visibility in AI
MCP is the standard everyone in AI agreed on. Anthropic invented it in late 2024, then OpenAI, Google, and Microsoft all adopted it within months. If you want AI assistants to take real actions on your site - book a table, place an order, look up an account - you ship an MCP server. Without one, agents are stuck reading your pages instead of using your service. The early-adopter window is wide open. Most sites have no MCP endpoint yet, so the few that do are getting first-class integration into agent tool catalogues. A restaurant with an MCP server can be booked by Claude directly; one without can only be described in a summary. Within a couple of years this will be table stakes for any business that wants agentic traffic to convert into transactions.
Pass criteria at a glance
| Criterion | Passes when |
|---|---|
| Valid MCP InitializeResult or advertised endpoint. |
How we test it
We POST a JSON-RPC `initialize` message to the conventional MCP paths with an Accept header of `application/json, text/event-stream`. A passing server replies with a JSON-RPC result, a `Mcp-Session-Id` header, and an `MCP-Protocol-Version` header. We also check your HTML for a `<link rel="mcp">` advertisement and your /llms.txt for an MCP endpoint reference, since the `.well-known/mcp` location is not yet standardised. Any of those three signals counts as a pass.
Show technical detection method
POST JSON-RPC initialize with Accept: application/json, text/event-stream; pass if response is application/json with JSON-RPC result and Mcp-Session-Id header, OR <link rel=mcp> in HTML, OR MCP-Protocol-Version response header.
If your site fails: how to fix it
- Decide which of your service's capabilities should be agent-callable (search, lookup, create, book, etc.) - these become your MCP tools.
- Deploy an MCP server using an official SDK (Anthropic, Cloudflare Workers, Stripe, Vercel) and expose the Streamable HTTP transport at a stable URL.
- Advertise the endpoint with `<link rel="mcp" href="https://yoursite.com/mcp">` in your HTML head, and reference it from /llms.txt.
- Implement OAuth 2.1 if your tools require authentication - MCP delegates to standard OAuth discovery (see the MCP transports spec at https://modelcontextprotocol.io/specification/2025-06-18/basic/transports).
- Because the spec is still evolving, have a developer who can read the MCP spec implement this rather than relying on a no-code tool.
Quick facts
| Maturity | EMERGING |
|---|---|
| Weight | high |
| Category | Emerging Standards |
Primary sources
Related signals
Frequently asked questions
Is MCP a real standard or just an Anthropic thing?
It started at Anthropic in November 2024 but is now an open specification (modelcontextprotocol.io) adopted by OpenAI, Google DeepMind, Microsoft, GitHub, Cloudflare, Stripe, and dozens of others. It is the dominant agent-tool protocol in 2025 - the closest thing to a cross-vendor consensus the AI industry has produced.
Which AI systems honour an MCP server today?
Claude (desktop and API), ChatGPT (via the Apps SDK), Microsoft Copilot, Cursor, Windsurf, Zed, and most agent frameworks (LangChain, LlamaIndex, Vercel AI SDK). Coverage in consumer chat assistants is growing rapidly through 2025 and 2026.
Do I need MCP if I already have a public REST API?
Agents can technically use a REST API, but MCP gives them structured tool definitions, schemas, auth flows, and session state out of the box. Without MCP each agent vendor has to hand-build an integration; with MCP they discover and call your tools automatically.
Will this still matter in two years?
Yes. The protocol is on a path to becoming the agent equivalent of what REST became for web APIs. Even if details change, the underlying pattern - a discoverable tool endpoint per origin - is here to stay.
Run your own scan
Run a free scan and see how your site grades across all 155 AI-readiness signals.