Use the downpipes docs in your AI tools

There are three ways to give an AI tool real downpipes documentation instead of a guess: an MCP search server that quotes the docs back with a source link, a curated llms.txt index for one-shot context, and the OpenAPI spec for API-shaped questions. All three are read-only. None of them can act on your Cloudflare account or your downpipes deployment; they only retrieve and return published pages.

What does the MCP server do?

https://docs.downpipes.io/mcp exposes one tool, search_docs. It searches the published documentation and returns verbatim passages, each with the source page’s URL, so an agent can quote and cite a real page rather than paraphrase from memory. It does not generate answers itself; that is the calling tool’s job, using the passages as grounding.

The index behind it is a crawl of the published site, not a live read of it, so a page that changed in the last few hours can still come back with its previous text. If the search result and the cited page differ, follow the page: open the URL the tool returns and read the live page whenever the two might disagree.

The server is only live once the site’s search backing is configured. Before wiring it in, check https://docs.downpipes.io/.well-known/mcp/server-card.json. A 200 there means the server is enabled; a 404 means it is not, and the llms.txt index below is the fallback.

Command-line tools that accept a URL

Many terminal-based AI coding tools can add a remote MCP server directly, typically with a flag such as --transport http followed by the URL. The URL is the only downpipes-specific part; the command name and flags come from your tool’s own MCP documentation.

<your-tool> mcp add downpipes-docs --transport http https://docs.downpipes.io/mcp

Desktop apps that only support stdio

Some desktop apps’ local config file only understands stdio servers, so a bare url entry is not supported there. Wrap the URL in a stdio bridge instead:

{
  "mcpServers": {
    "downpipes-docs": {
      "command": "npx",
      "args": ["mcp-remote", "https://docs.downpipes.io/mcp"]
    }
  }
}

Check your app’s own connector settings first: some accept a URL through their UI even when their config file does not.

Cursor

Add this to mcp.json:

{
  "mcpServers": {
    "downpipes-docs": {
      "url": "https://docs.downpipes.io/mcp"
    }
  }
}

What is llms.txt for?

https://docs.downpipes.io/llms.txt is a plain-text index of every public docs page, grouped by section, each with a one-line description. It is built for a coding agent that loads it once as background context rather than calling a tool per question. Paste its contents, or the URL itself if your tool fetches links, into a session before asking about downpipes.

Where is the OpenAPI spec?

https://docs.downpipes.io/api/openapi/engine.json is the OpenAPI 3.1 description of the engine’s admin HTTP API: the method, path and tag of each route in the router’s admin switch. It does not list the sign-in routes under /admin/auth/, /admin/oidc/ and /admin/saml/, which the router dispatches by prefix. The sign-in flows page lists those routes. The spec is a snapshot of one engine version, so it does not prove that a deployed engine matches the spec. It covers routes, not full request or response schemas, so read the API overview and the per-endpoint reference pages alongside it.

Does any of this generate answers about downpipes?

No. search_docs and llms.txt both return retrieval only: real passages, from the published docs, with a source URL attached. Neither one writes a summary or an opinion. If an AI tool answers a question about downpipes in its own words, that answer came from the tool’s own model reasoning over the retrieved passages, not from downpipes. Treat the quoted passage and its source link as the fact, and treat the surrounding answer as the tool’s interpretation of it.

Last updated 25 September 2026.