flow

For agents

Made for people.
Readable by agents.

Every page people read is also available as Markdown, with a search API, an ask API, and a read-only MCP server beside it. Nothing needs a login.

Discover, retrieve, understand

  • 01 / Discover

    Discover

    llms.txt, a sitemap, OpenAPI, and MCP discovery documents point agents to the right skill or guide.

    Read llms.txt
  • 02 / Retrieve

    Retrieve

    Stable URLs, Markdown for every page, and structured search results with canonical links and excerpts.

    Try the search API
  • 03 / Understand

    Understand

    Answers grounded in Flow's own sources, with numbered citations that link back to the canonical page.

    See the ask example

Public contracts

Clear capabilities.
Small surface area.

  • GET /llms.txt

    Curated entry point: every skill, guide, capability, article, decision, starter, and interface, with Markdown links.

  • GET /llms-full.txt

    Every guide, capability, article, and skill as one Markdown document, with source paths and the content commit.

  • GET /api/search?q=

    Ranked results with canonical URLs and excerpts. Optional limit (1–25) and kind (skill, doc, decision, starter, capability, article).

  • POST /api/ask

    Source-grounded answer with numbered citations. Send {"question": "…"}; add ?stream=1 to stream text.

  • MCP /api/mcp

    Streamable HTTP, stateless, read-only. Tools: search_flow, list_skills, get_skill, get_doc, ask_flow. Also at /mcp.

  • GET /<page>.md

    Markdown for any page: append .md (home is /index.md) or send Accept: text/markdown.

  • GET /openapi.json

    OpenAPI 3.1 for search, ask, and the MCP endpoint, including the error schema.

  • GET /.well-known/mcp.json

    MCP discovery, with the server card at /.well-known/mcp/server-card.json and an AI catalog.

  • GET /sitemap.xml

    Every page on the site, for crawlers and agents that walk the whole catalog.

All interfaces are live, public, and read-only. Errors are JSON with code, error, resolution, and type; rate limits answer 429 with Retry-After.

Try it

Copy, paste, read.

curl -s https://flow-skills-md.vercel.app/llms.txt
curl -s 'https://flow-skills-md.vercel.app/api/search?q=ship&limit=3'
curl -s -H 'Accept: text/markdown' https://flow-skills-md.vercel.app/skills/build
curl -s https://flow-skills-md.vercel.app/api/ask -H 'Content-Type: application/json' -d '{"question":"How do I ship a change?"}'
MCP client config
{
  "mcpServers": {
    "flow": {
      "type": "http",
      "url": "https://flow-skills-md.vercel.app/api/mcp"
    }
  }
}
MCP tool call
curl -s https://flow-skills-md.vercel.app/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_skill","arguments":{"name":"build"}}}'

Verify the result

Agent-ready is observable.

Check the HTML, Markdown responses, API discovery, structured errors, and real agent journeys yourself, and report what you measure.

npx is-agentic flow-skills-md.vercel.app

Jump to, or try a question