How can a coding assistant fetch the Tares docs as plain Markdown, and does an Accept header work?

Tares serves every docs page as plain Markdown when you append .md to the page URL. Start from the index at /tares/llms.txt, and use the all-in-one file /tares/llms-full.txt only when a few pages are not enough. An Accept: text/markdown header does not work, because the server answers with HTML, so use the .md suffix.

How it works

Tares publishes its docs in a form a coding assistant can read without scraping rendered pages. The AI resources page lists the endpoints:

• /tares/llms.txt is the Tares index. It lists every page with a one-line description, in reading order. This is the place to start.
• /tares/llms-full.txt holds all Tares pages in one file. It is large, so reach for it only when one or two pages are not enough.
• Any page URL plus .md returns that page as Markdown. For example, the Prometheus connector page becomes /tares/connectors/prometheus.md. The docs root is /tares.md.
• The site also has a broader /llms.txt that covers other products. You do not need it for Tares work.

The Markdown is the same content as the rendered page, without the navigation. That makes it the better input for an assistant than the HTML.

Content negotiation is not supported. If you send an Accept: text/markdown header to a normal page URL, the server still answers with HTML. The .md suffix is the only way to get Markdown for a page.

Set it up

Give your assistant this retrieval order, whether you are Claude Code, Codex, Cursor or an agent of your own.

1. Fetch /tares/llms.txt and read the one-line descriptions.
2. Pick the one or two pages that match the task and fetch their .md URLs. The task router on the AI resources page maps common questions to pages. For example, installation questions go to the quickstart, then the CLI reference. Questions about adding a source go to the connectors index, then the connector’s own page.
3. Fetch /tares/llms-full.txt only if the answer spans many pages.
4. Do not send an Accept header to ask for Markdown, and do not scrape the HTML pages. Append .md instead.
5. Treat llms.txt as a map, not as an answer. The answer is on the pages it points to.
6. Put the rule in a system prompt or a project instruction file, so the assistant follows it every time:

/tares/llms.txt                      first, the index
/tares/connectors/prometheus.md      then the matching page, as Markdown
/tares.md                            the docs root

7. Tell the assistant to cite the page URL it used, and to take connector config keys only from that connector’s page. If a key is not listed there, it should say so and not invent one.
8. If the running instance disagrees with the docs, trust the instance and say the docs may be behind. Check the instance with:

tares status

If you are building an assistant that watches systems and answers incident questions, the AI SRE guide is the page to fetch first after the index.

What you get

• Clean Markdown for any Tares docs page from a single URL change, with no navigation around it.
• A small, ordered entry point in /tares/llms.txt, so the assistant reads two pages and not the whole site.
• One big file, /tares/llms-full.txt, for the cases that span many pages.
• Answers that stay close to the docs: commands, flags and connector keys come from the pages, and the assistant cites the URL it used.
• No surprises about headers. You know that Accept: text/markdown returns HTML, so you do not build logic around it.

When something else fits

• If the answer spans many pages, /tares/llms-full.txt saves you from fetching them one by one, at the cost of a large file.
• If you need the live state of a running Tares and not the docs, ask the instance directly with tares status. The docs can be behind a running instance.

Related