Research · Get found by agents · 13 October 2026
How to make your site and tool ready for agents
An agent landing on your site wants to know what you do, how to call you, what it costs and whether you're up. We counted how many of the 966 tools we list answer each of those, and what answering them did for us.
What agents look for, and who has it
| What an agent looks for | Tools we list that have it |
|---|---|
| A changelog | 800 of 966 (83%) |
| A status page | 725 (75%) |
| An llms.txt | 717 (74%) |
| A way to start without talking to sales (free, freemium or pay as you go) | 716 (74%) |
| An OpenAPI description | 523 (54%) |
| An MCP server | 487 (50%) |
| A valid security.txt | 266 (28%) |
| A place in the official MCP registry | 162 (17%) |
All of these come from the listings in our directory [1]. None of them is hard, and the bottom of the table is where most tools lose an agent.
Start with llms.txt
llms.txt is a Markdown file at /llms.txt that tells an agent what a site is and where its useful pages are. Jeremy Howard proposed it on 3 September 2024 [2]. The only part it requires is a heading with the name of the site, and after that come a one-paragraph summary, any notes, and sections of links, each with a line saying what's there [2].
Put the pages that matter to an agent in it, the API reference, authentication, rate limits, pricing and errors, rather than the marketing pages. If prices or limits are in it, say when they were checked. Three in four tools we list already publish one, and the share drops with the grade, from 85% of tools graded BB down to 38% of those graded F [1].
Then give every page a Markdown copy
An agent that fetches HTML spends its context on navigation, scripts and cookie banners. On our site every page has a Markdown copy and a JSON copy, 14,874 Markdown files in the last build, and a request with Accept: text/markdown gets the Markdown without having to know the address [3].
We built it that way because agents don't always read the file you leave for them. Even with all of that in place, agents fetched the HTML of a page 1,085 times in our first week when the Markdown was right there [4].
Describe the API, and keep the boring pages
Publish an OpenAPI description, because any agent that can make an HTTP request can use it, and the previous post has why we'd do that before an MCP server. Keep a status page and a dated changelog, because an agent deciding whether to rely on you checks whether you're up and whether you change things without warning. And publish a security.txt, the cheapest item on the list and the one only 28% of tools have [1].
What it did for us
We built Anchor Terminal agent first, with an llms.txt, a Markdown and JSON copy of every page and an MCP server, and we think that's a big part of why we were found so quickly. Six days after we went live, crawlers made 71,359 requests to the site in a single day [4].
And claim your listing
If your tool is in our directory, claim it. It's free, it's proved by a DNS record, an HTTP challenge or a commit, it lets you correct the facts on the listing, and it never changes a score [5]. If it isn't listed yet, the same page takes a request.
References
Read on 10 October 2026.
- [1]Anchor Terminal, the directory, 10 October 2026. Source for the share of tools with each item and llms.txt by grade. Each listing links the pages it was graded from. anchorterminal.com
- [2]Jeremy Howard, The /llms.txt file, llmstxt.org, 3 September 2024. Source for the proposal, its date and the format. llmstxt.org
- [3]Anchor Terminal, llms.txt and the build of 10 October 2026. Source for our Markdown and JSON copies and the Accept header. anchorterminal.com
- [4]Anchor Terminal server logs, 1 to 8 October 2026, not published. Source for the HTML reads by agents and the crawler requests.
- [5]Anchor Terminal, For builders, Claiming. Source for free claiming, its proofs and claims never changing a score. anchorterminal.com