About & methodology
Personal AI agents are now shopping, booking and logging in for people, and companies are deciding whether to let them. Agents Allowed tracks those decisions in one place, with sources, so people and agents know what to expect before they try.
What the access levels mean
| Status | Meaning |
|---|---|
| Open access | A live official route, or the website itself, lets agents finish the main task from start to finish. |
| Partial access | Agents can do some tasks but not finish the main one (for example, checkout is handed back to a person), or the site's verification checks can stop agents. |
| Announced access | The brand has announced an official way in for agents, but we haven't confirmed it's live yet. |
| No current access | Agents can't currently use the site for the listed tasks, under its terms or technical measures, per the sources. |
Recommended next step
Each listing suggests what an agent should do next. It's based on public sources and isn't permission; the website's own terms always come first.
| Next step | When we show it | What it means |
|---|---|---|
| Use official route | Open or Partial access, and the brand has a live official way in for agents | Use that official route instead of the website. It's the most reliable option. |
| Proceed with caution | Open or Partial access, but no live official route yet | Some tasks may work on the website, but expect "are you human?" checks. If one appears, hand the task back to the person. |
| Proceed with caution | Announced access | The official route isn't confirmed live yet, and the website may not work for agents. If the agent gets blocked, hand the task back to the person. |
| Do not proceed | No current access | Don't try these tasks on the website for now. Tell the person, or use a listed official route if one covers the task. |
| Ask the user | We don't have a verified record for the site | Follow the site's terms and robots.txt, and check with the person before acting. |
Evidence levels
| Evidence | Meaning |
|---|---|
brand_statement | company statement, spokesperson, executive or terms |
official_docs | Official documentation or announcement from the brand or platform partner. |
credible_reporting | Reported by a reputable outlet (named reporting or a documented hands-on test by the outlet). |
own_test | Our own documented, dated test with a real agent. |
How we check
- Sources first. Every record links to dated sources: brand statements, official docs, and reputable reporting. We never assign a status without a source.
- Verified only. We publish a record only when it rests on a brand statement or official documentation, reporting by a reputable outlet (named and dated, including an outlet's documented hands-on test), or our own documented, dated test with a real agent. User reports and social posts alone are never published: they go to an internal review queue until one of those sources confirms them.
- Submissions are reviewed, never auto-published. Anything sent through Submit an update, the API or the MCP tool goes to a human review queue first.
- Official routes are listed when a brand has a live, announced or planned way in for agents: Google's Universal Commerce Protocol (UCP), OpenAI/Stripe's Agentic Commerce Protocol (ACP), MCP servers, Visa's Trusted Agent Protocol, A2A / Personal Agent Protocol, partner apps (e.g. in ChatGPT or Muse), or paid data APIs.
- Dates. "Last checked" is when we last reviewed the record. For documentation pages without a publish date, the source date is the date we read it. Every change is logged in the changes feed.
- Not legal advice. A listing describes observed behaviour and public statements. Always follow a site's own terms.
Coming next: our own documented hands-on tests with real agents (agent, date, steps, result, screenshot), so records can cite own_test evidence, and a regular re-check schedule for every brand.
Data license
All rights reserved. Free to query and to use results with attribution under the Terms of Use; an open data license may be added later. See Terms, section 3.
Corrections
Brands and readers can request a correction by emailing contact@agentsallowed.io (the update form opens soon). A person reviews it against sources; if it checks out we update the record, its "last checked" date and its public change history. Brands can also ask us to review how they're described. See the disclaimer.
Sponsorship
We don't take sponsors yet. If we do, sponsors will never influence a brand's status, and sponsored placements will always be labelled.