llms.txt & Markdown docs

Everything on this docs site exists twice: once as the HTML you're reading, and once as Markdown built for machine consumption. Agents don't need a rendered page — they need clean, link-rich text they can index, retrieve from, or load whole into context. This page explains the three forms that text takes and how they relate.

The corpus

/llms.txt — the index: a flat map of every page with a one-line summary and a link each. This is the llms.txt convention — small enough to always fit in context, ideal as a starting point an agent follows links from. Use it when the agent retrieves on demand.
/llms-full.txt — the full corpus: every doc page concatenated as Markdown in one file. Use it when you'd rather hand the agent everything up front — a long-context session, a RAG ingestion job, or a project-level knowledge file.
The per-page Markdown twin — every individual page is also available as raw Markdown, so an agent (or you) can grab exactly one page without the rest of the corpus.

All three are generated from the same sources as the HTML site and regenerated with every deploy, so they never lag the published docs.

The Copy-page affordance

You rarely need to remember those URLs. Every page across the Guides, API reference, SDK, and knowledge-base sections carries a Copy page menu that hands the machine-readable forms to your tools directly:

Copy page as Markdown — the current page's Markdown twin, ready to paste into a chat or a context file.
Open llms-full.txt — the full corpus, when one page isn't enough.
Copy MCP command — the bird mcp server command from the MCP server guide, so the page your agent is reading can also wire up the tools to act on it.
Connect to Cursor / Connect to VS Code — one-click MCP setup for the two most common agentic editors.

The dashboard mirrors the same affordances: the email onboarding page's Copy for AI menu offers the identical Markdown, corpus, and MCP shortcuts, so the path from "I'm in the dashboard" to "my agent knows Bird" is one click from either side.

The API reference never drifts

The API reference section is generated from Bird's OpenAPI specification — the same spec the API itself is validated against. Its Markdown twin is generated from that same source, which means the machine-readable reference can't drift from the live API surface: when an endpoint changes, the spec changes, and the HTML, the Markdown, and llms-full.txt all regenerate together. An agent reading the corpus is reading the contract, not a hand-maintained copy of it. Start at the API reference introduction to see how the reference is organized.

Choosing a form

You want the agent to…	Hand it…
Discover what exists and fetch pages on demand	/llms.txt
Have the entire docs corpus in context or in a RAG store	/llms-full.txt
Understand one topic precisely	That page's Markdown via Copy page
Act on Bird, not just read about it	The MCP server — docs alone can't send an email

For the hand-held, end-to-end setup — MCP server installed, editor connected, corpus wired in — follow AI onboarding.