SecondDoor

Making your product discoverable to AI agents

MCP manifests, agent-aware robots rules, and what discoverable actually means to a machine.

Publish an interface, then advertise it

An MCP server gives an assistant a set of typed tools instead of a website to scrape. Serving a manifest at /.well-known/mcp.json is what makes it findable without someone being told it exists.

Most products that have an API have most of this already. What is missing is the declaration: the machine-readable statement that the interface exists and where it lives.

Check what your robots.txt says about assistants

Rules blocking GPTBot, ClaudeBot, PerplexityBot and their peers are frequently inherited from a template nobody revisited. They are worth an explicit decision rather than a default.

If you want to be cited by assistants, name those crawlers and allow them. If you do not, that is a defensible position, but it should be one you chose.

Questions

Do we need MCP if we already have a REST API?
A documented REST API with an OpenAPI spec covers most of the same ground. MCP is more discoverable and needs no client written for it. The spec matters more than the protocol.
Does blocking AI crawlers protect our content?
It prevents citation as much as it prevents training. Decide which of those you were trying to stop, because the same line does both.
How agent-aware instrumentation is scoredCheck your own site
Making your product discoverable to AI agents, SecondDoor