Write a review

Blog · Opinion · 22 October 2026

The tools that already let agents in

Some tools have already fixed the part of agent use that matters most, the front door. An agent can start using 22 of the tools we've checked without a person, and the way they do it is simple enough to copy.

Preview This post is scheduled for 22 October 2026 and may change before then. It isn't on the blog yet.

By Vlad Cealicu1,230 words5 min read

I've spent a week writing about what's broken, personal agents that can't finish an errand, MCP and benchmark bans. This one is about the tools that got it right.

The hardest part for an agent is usually the first minute. Before it can call anything it needs an account, and accounts are made for people, a form, an email to confirm, a card, a dashboard to copy a key from. We've gone through that first minute step by step for 103 listings so far, starting with the ones graded BB or better. For 22 of them an agent can start on its own (21 products, since Telnyx is listed twice, for messaging and for voice) [1].

1,112tools we list and grade
174agent-ready, a grade of BB or better
22an agent can start using alone, of the 103 we've checked
Anchor Terminal, 16 October 2026. Source [1] [2].

How do they let an agent in?

Four ways, and none of them is complicated.

Setup checks, 14 October 2026How the 22 listings let an agent start alone

Listings

  1. A sign-up API made for agents (Telnyx counted twice)7
  2. Free calls with no key6
  3. Nothing to sign up for5
  4. Pay at the door4

Bars start at zero. Values are labelled in each panel.

What's a sign-up made for agents?

My favourite is ZenRows. An empty POST to its agent sign-up endpoint returns a working key on the Free plan and a link a person can open later to take ownership of the account [3]. The agent starts now, and the person catches up when they like. No form, no card, no waiting.

ZenRowsFrom nothing to a working key
  1. POST /api/agent/signupAgent alone

    An empty body, no account and no key [3]

  2. Use the keyAgent alone

    The Free plan, 5,000 credits a month, no card [3]

  3. Open the claim linkWhenever they like

    A person takes ownership of the account, later [3]

  • Agent alone
  • Whenever they like

AgentMail does something similar for email. A sign-up call returns an API key and an inbox, and a person reads the agent a 6-digit code before it can send to anyone but them [4]. Didit, an identity checker, registers through an API with an inbox the agent can read and a code it sends there [5]. Telnyx has a bot challenge and a bot sign-up that give an agent a trial account, and documents topping up credit over x402 where the account is eligible [6]. monday.com goes the furthest on paper. Its auth.md describes a sign-up API that asks the agent to solve a captcha made for agents and returns an API token [7], the opposite of every bot wall I hit trying to book a flight. Its own sign-up guide for agents still says "There is no API that provisions one for you" [7], so one of the two is out of date. Composio's CLI has composio login --agent, which makes an agent its own account and key [8].

What about free calls with no key?

Six tools let an agent call them with no key at all, inside a rate limit. Context7's hosted server answers with no key at low rate limits [9]. Tavily takes a header, X-Tavily-Access-Mode: keyless, for search and extract [10]. Firecrawl's hosted MCP server has a rate-limited keyless tier, Parallel's search server works with no key at lower limits, and You.com's gives 100 searches a day with no account [11]. Tempo's read endpoints answer without a key, and when an agent goes over the limit they answer 402 so it can pay for the next request over MPP [12]. That last one is how I'd want every free tier to end.

A key exists to tell the provider who to bill and who to throttle. A rate limit per IP does the throttling, and a payment does the billing. For a lot of tools the key is just in the way.

And paying at the door?

Four tools let an agent pay instead of signing up. Browserbase sells browser sessions by the hour over x402, with no account or key. Spider answers a scrape with a 402, the agent pays in USDC and gets the data. Browser Use makes a wallet-linked account from the first payment. Apify sells a prepaid token over x402 from $1, still marked experimental [13]. This is the payment channels argument from the other side. A paid call needs no account, so it needs no form.

And nothing to sign up for?

Five are open source or source-available and run where the agent runs, Selenium, Chrome DevTools MCP, Robot Framework, Arize Phoenix and Microsoft's execution containers [14]. Install, run, no account. It's easy to forget that the oldest way to let an agent in still works.

Does the front door make a difference?

For the person too. Even the normal route, with a person signing up, is quicker for these. We estimate a median of 2 minutes of a person's time to set one of the 22 up, against 10 minutes and three steps for the other 81, and some of those, like Mastercard Agent Pay, have no self-serve route at all. 17 of the 22 need no card (the ones you pay at the door need a funded wallet instead) [1].

2 mina person's time to set up one of the 22, median
10 minfor the other 81
17 of 22need no card
Setup checks of 103 listings, 14 to 16 October 2026, estimated. Source [1].

The surprise is who isn't on the list. Of the 18 tools with an A or AA grade, we've checked setup for 16, and an agent can start alone with none of them [1] [2]. Some of that is fair. Google Calendar, Google Drive, Stripe and Twilio act on a person's own data or money, and a person should be in that first step. The rest are documented, reliable and well priced, and they still start with a person in a dashboard. For them the front door is the step they haven't taken, and the 22 show it's a small one.

Our panel's reviews point the same way. These are desk reviews, written from the same public evidence we grade from, so they're a consistency check and not independent proof. The share with four or five stars falls from 50 per cent for tools graded A to 6 per cent for F (AA is a single tool, so I've left it out) [15].

Panel desk reviews, October 2026Reviews with four or five stars, by grade, per cent

Per cent

  1. A50
  2. BB42
  3. B33
  4. C27
  5. D20
  6. E18
  7. F6

Bars start at zero. Values are labelled in each panel.

Which areas are doing best?

Agent runtimes, developer tools and payments, where a third or close to it is agent-ready. Design tools have none yet [2].

Anchor Terminal, 16 October 2026Agent-ready tools by area, per cent

Per cent

  1. Agent runtime34
  2. Developer & infrastructure29
  3. Payments & protocols28
  4. Voice & speech24
  5. Agent frameworks20
  6. Everyday APIs17
  7. Communication17
  8. Web, data & documents13
  9. Models & inference12
  10. Work & business apps11
  11. Domain data4
  12. Content generation3
  13. Design & diagrams0

Bars start at zero. Values are labelled in each panel.

What's already there?

More than I expected. 800 of the 1,112 listings point to an llms.txt (726 different files, since some companies have several listings), 722 publish prices per unit and 552 point to an OpenAPI description (512 different files) [2]. Of the 674 vendors whose terms we've read, 359 have no clause stopping you from benchmarking them or building something that competes [16]. Most of the work for agents is done. The first minute is what's left.

If you run a tool, copy ZenRows. One endpoint that returns a key, a link for a person to claim it later, and a free tier with a limit. If you've built something like that, tell us and we'll check it. What makes a tool agent-ready beyond the door is in how to make your tool agent-ready.

References

Read on 16 October 2026.

  1. [1]Anchor Terminal, setup checks for 103 listings, 14 to 16 October 2026, 99 of them graded BB or better. Each records whether an agent can start alone, how, a person's steps, an estimate of a person's minutes, and whether a card is needed. Source for the 22, the four ways in, the medians of 2 and 10 minutes, the cards and the A-grade tools. anchorterminal.com
  2. [2]Anchor Terminal, the directory API, 16 October 2026. Source for the 1,112 listings, their grades, the 174 agent-ready, the 18 graded A or AA, the shares by area, and the llms.txt, price and OpenAPI counts. anchorterminal.com
  3. [3]Anchor Terminal, the ZenRows listing. Source for the agent sign-up endpoint, the claim link and the Free plan. anchorterminal.com
  4. [4]Anchor Terminal, the AgentMail listing. Source for the sign-up call returning a key and an inbox, and the 6-digit code. anchorterminal.com
  5. [5]Anchor Terminal, the Didit listing. Source for the registration API and the emailed code. anchorterminal.com
  6. [6]Anchor Terminal, the Telnyx listing. Source for the bot challenge, the bot sign-up, the trial account and topping up credit over x402 where the account is eligible. anchorterminal.com
  7. [7]monday.com, auth.md and the signup-for-agents skill at /.well-known/agent-skills/, read on 16 October 2026. Source for the sign-up API at signup-logic.monday.com with its captcha for agents and the api_token it returns, and for the skill saying no API provisions an account. monday.com
  8. [8]Anchor Terminal, the Composio listing. Source for composio login --agent. anchorterminal.com
  9. [9]Anchor Terminal, the Context7 listing. Source for the hosted server answering with no key at low rate limits. anchorterminal.com
  10. [10]Anchor Terminal, the Tavily listing. Source for the keyless header. anchorterminal.com
  11. [11]Anchor Terminal, the Firecrawl, Parallel and You.com listings. Source for their keyless tiers and You.com's 100 searches a day. anchorterminal.com
  12. [12]Anchor Terminal, the Tempo listing. Source for read endpoints with no key and the 402 over the limit. anchorterminal.com
  13. [13]Anchor Terminal, the x402 page and the Browserbase, Spider, Browser Use and Apify listings. Source for paying per session, per call, for an account or for a prepaid token. anchorterminal.com
  14. [14]Anchor Terminal, the Selenium, Chrome DevTools MCP, Robot Framework, Arize Phoenix and Microsoft execution containers listings. Source for running with no account or key. anchorterminal.com
  15. [15]Anchor Terminal, the panel's reviews, October 2026, all desk reviews from public evidence. Source for the share of four- and five-star reviews by the tool's grade. anchorterminal.com
  16. [16]Anchor Terminal, vendors' terms, read in October 2026. Source for the 674 vendors and the 315 with a restrictive clause. anchorterminal.com

Your experience matters

Used a tool in the Terminal?

Tell people and their agents what worked and what didn't. Write a review of a tool you've used and confirm it by email or wallet.

Write a reviewHow reviews work