Agent Discovery Files

MDflow publishes machine-readable files so AI agents and tools can discover what it offers and how to operate it — without a human pasting documentation into a prompt.

The files

File	What it is
llms.txt	A discovery index pointing agents to the docs, pricing, API and MCP server (see 03 MCP Servers)
docs.md	A self-contained control manual — authentication, every MCP tool, the REST endpoints, and client setup, in one page
pricing.md	Machine-readable pricing, so an agent can tell what a workspace's plan allows (see 04 Free and Pro Plans)
openapi.json	The OpenAPI 3.1 specification for the REST API (see 02 HTTP API)
/.well-known/agent-card.json	An A2A (agent-to-agent) discovery beacon describing MDflow's interfaces, security scheme and skills

All are listed in the sitemap and served with explicit markdown and JSON content types.

docs.md is the one to hand an agent that needs to operate MDflow. It is deliberately one self-contained page rather than a site to crawl.

Raw markdown for any shared page

Append .md to a share URL (see 01 Public Share Links) and get the document as plain markdown:

https://mdflow.cz/share/{slug}      → the rendered page
https://mdflow.cz/share/{slug}.md   → the raw markdown

Each raw response begins with YAML frontmatter — title, canonical URL, markdown URL, visibility — so a tool can cite and read it in a single request, with no HTML parsing and no guessing at the canonical URL.

This works for shared collections too.

Why this matters to you

Two practical consequences for anything you publish from MDflow:

Models read your shared documents correctly. A shared MDflow document is already in the format a language model wants — clean markdown with declared metadata, not markup wrapped around navigation and cookie banners. There is nothing to extract and little to get wrong.

Your published documents are citable. The frontmatter carries the canonical URL, so a model quoting your document has the right link to attribute it to.

If you publish reference material you actually want models to represent accurately, this is the mechanism.

Share pages are not indexed

Worth being clear about the boundary: share pages are marked noindex. They follow an unlisted link model — discoverable by anyone you give the URL to, not by search engines crawling for them.

So the discovery surface describes MDflow the product, not your content. An agent can learn how to operate MDflow from these files; it cannot use them to find your documents. Your shares are reachable only by their link.

The design idea

MDflow's premise is that the same document should serve a person and a machine equally well, with no export step between them.

The discovery files extend that to the product: a capable agent can arrive at mdflow.cz knowing nothing, read llms.txt, follow it to docs.md, learn every tool and endpoint, check pricing.md to see what the plan permits, and start working — without a human explaining anything.