The website
Give it to your agent
Three ways to hand content from Context Atlas to a coding agent - a prompt, a link or Markdown - and when to use each.
The prompt from /agent
/agent has one sentence to paste into Claude Code, Codex, Cursor or any agent with web access. Pick the mode first:
- I have a project: run it inside your repository. The agent detects the stack, tells you what it found, then loads the matching sources.
- I'm starting something new: describe what you want to build. The agent proposes a stack, explains the trade-offs and sets it up once you agree.
The prompt only points to /llms.txt. The real instructions live there, so they stay current without you changing the prompt.
Read https://contextatlas.dev/llms.txt and follow its instructions for an existing project: detect the stack of this repository, tell me what you found, then load the matching sources and follow their best practices.
Links
Agents fetch plain-text URLs reliably. Paste a link into the chat, or put it in your project's AGENTS.md or CLAUDE.md so every session sees it.
| To give the agent… | Paste |
|---|---|
| The whole catalog with instructions | https://contextatlas.dev/llms.txt |
| The sources of one tool | https://contextatlas.dev/llms/<id>.txt, such as https://contextatlas.dev/llms/stripe.txt |
| The catalog and all docs at once | https://contextatlas.dev/llms-full.txt |
| One docs page | The page URL with .md, such as https://contextatlas.dev/docs/mcp.md |
| Structured data for a script | https://contextatlas.dev/atlas.json or the JSON API |
Prefer the topic files over the whole catalog when you know your stack. They are small and leave room in the context for the vendor docs themselves.
A line for AGENTS.md:
Before working with a tool from our stack, fetch its sources from https://contextatlas.dev/llms/<id>.txt (ids: nextjs, drizzle, vercel) and follow them.
Markdown
Every docs page has a Markdown version: add .md to its URL, or use the link at the bottom of the page. Every docs page also announces it with <link rel="alternate" type="text/markdown">, so agents that read a page's head find it on their own.
MCP instead
If your agent supports MCP, connecting the MCP server is usually better than links: the agent searches and fetches only what it needs, without you choosing files.