Can an AI agent use Agentgates?
Agentgates sells virtual machines, AI model access, phone numbers, email inboxes and a directory of paid APIs, all for AI agents.
Rank among the 246 products we drove
- Time to get in
- tied #2
- <1minutes
- Getting in on its own
- Steps on its own
- tied #86
- 2/ 5
- Find to Leave, no person
Partly. Our agent got into Agentgates on its own without signing up and used its API, MCP server and CLI. It offers agents a way to pay: x402.
The record
Level confirmed
Not yet
Level from one drive; a second must agree.
- Got in by
- No key neededUnder a minute, no account
- Got as far as
- Leave: can't delete its accountCan pay on its own
Connects by
MCP, API, CLI
Agent payments
x402; our agent used it
Partly · 32 doc pages read
Docs with gaps: no getting-started page; has llms.txt, an OpenAPI and an MCP server. The level doesn't rest on Find: it counts getting in, using, paying and leaving.
On its own · <1 minute
Got in without signing up.
Not measured
Work not reached.
On its own · 1 way for agents to pay
Valid x402 payment challenge: $0.002 a call.
Partly
Nothing to cancel. Its account stays.
- Partly
- On its own
- Not measured
To reach L3: Let a trial or free plan reach the core features.
Checked 18 Nov 2026: MCP server asks for a sign-in, x402 payment challenge valid, llms.txt and OpenAPI unchanged since the check before.
Find
How an agent connects
MCP, API, OpenAPI, CLI, agent payments and llms.txt found.
MCP server
3 of 4 testedOfficial server. Keyless compute MCP (3 tools when driven, answers the 25 Nov 2025 MCP spec); OAuth wallet MCP behind a person sign-in. Sign‑in needed to see its tools.
4 tools mapped in our drive
API
59 of 118 testedREST at api.agentgates.ai; OpenAPI 3.1.0 for compute, directory and inference; account doors in prose.
118 operations mapped in our drive
OpenAPI
YesA machine-readable description of the API.
CLI
30 of 45 testednpm @agentgates/cli 0.88.0.
45 commands mapped in our drive
SDK
None foundllms.txt
YesAgent payments
Yesx402 (Base, Polygon, Solana).
Walked to the last click ("PAYMENT-SIGNATURE retry") and stopped: compute.starter 1h. No CAPTCHA at checkout. It offers x402 payment.
Docs
Some gapsCovers auth and error codes.
Code examples on only some pages.
No getting started.
Describes each operation only in part: completeness 50 of 100.
2 of 12 documented ways to do the core work were tried, and 1 of them worked.
32 of 38 pages read
Get in
Getting in
Our agent tried 3 of 4 ways in; 2 worked without a person.
The way that worked
No key neededunder a minute
- An inboxNot needed
- A browserNot needed
- A personNot needed
- Agent on a serverAPI, MCP, CLIPartlySigns up, gets an API key, makes its first API call and does a first task. The core work wasn't measured in this drive. It can't delete its account.
- Assistant in its own browserLike ChatGPT agentPartlySigns up and does a first task. The core work wasn't measured in this drive. It can't delete its account.
- Signed agentKnown to CloudflareYesThe site opens.
Every way in
| Way in | What happened | Why |
|---|---|---|
| No key needed | Worked | |
| Key by API | Worked | |
| Pay per call | Not reached | |
| Sign-in for apps (OAuth) | Stopped | The consent page needs an account |
Use
What an agent can do with Agentgates
The core task done, and 6 of 9 others.
Core tasks
- done there
- tried, not done
- offered, not tried
| Task | Result | Through |
|---|---|---|
| Search the x402 directory of paid APIs | Done | API: done there |
| Rent a confidential or plain virtual machine | Paid plan, not counted | API: tried, not done |
| Destroy a virtual machine | Paid plan, not counted | API: tried, not done |
| Run an LLM request on a confidential or open modelNot run: each request needs a payment and our wallet held $0. | Paid plan, not counted | API: tried, not done |
| Get a phone number of your own | Paid plan, not counted | API: tried, not done |
| Have the AI operator place a phone call for a taskNot run: the phone plan draws $3 from a balance our wallet lacked. | Paid plan, not counted | API: tried, not doneCLI: offered, not tried |
| Create an email inbox for an agent | Paid plan, not counted | API: tried, not doneCLI: offered, not tried |
| Send an email from an agent inbox | Paid plan, not counted | API: tried, not done |
| Read the mail in an agent inbox | Paid plan, not counted | API: tried, not doneCLI: offered, not tried |
Surface audit: 175 documented operations: 97 run, 45 blocked by sandbox or plan limits and 33 left out because they spend money or reach real people. Of the 97 run, 80 passed and 17 didn't; 3 behaved differently from the docs. By surface, 59 on the API, 30 in the CLI, 3 on the MCP server and 5 elsewhere, such as sign-in and webhooks.
Holds up
82of 100
- Reliability100
- Handles mistakes64
The mean of the two.
Docs match
82of 100
- Docs accuracy95
- Docs completeness50
Docs accuracy counts 70%, completeness 30%.
Surface tested
56of 100
80 of 142 passed: the 175 documented operations less the 33 that spend money or reach real people
What it can't do, and why
- Rent or destroy a virtual machine Not run: it needs a payment and our test wallet held $0.
- Run an AI request Not run: each request needs a payment and our wallet held $0.
- Get a phone number, send texts or place AI calls Not run: the phone plan draws $3 from a balance our wallet lacked.
- Create an inbox, send or read mail Not run: an inbox needs a funded wallet; the CLI's create command only prints a web link.
- Use the sign-in protected MCP server Its sign-in is for people and does not accept wallet accounts.
6 of 9 other tasks done
- Create a voice persona for callsDone
- Read one directory listing and its reviewsDone
- Store a site login in the vaultDone
- Save a file to agent memoryDone
- Read what keys spentDone
- Prove an email on the accountDone
- Renew a virtual machine for more hoursPaid plan
- Stop, start or restart a virtual machinePaid plan
- Resize a virtual machinePaid plan
- Replace the workload on a running virtual machinePaid plan
- Read a virtual machine's status and attestation proofPaid plan
- Wait for a sign-up code sent to an inboxPaid plan
- Release an email inboxPaid plan
- Read texts received on a numberPaid plan
- Send an SMS from a numberPaid plan
- Wait for a verification code texted to a numberPaid plan
- Place a call from your own numberPaid plan
- Buy a payphone call with no accountPaid plan
- Set how a number answers inbound callsPaid plan
- Read call transcripts and recordingsPaid plan
- Send a browser agent to a site with a taskPaid plan
- Top up the prepaid balancePaid plan
- List your own paid API in the directoryBlocked
- Review a paid API you usedNot tried
- Open a support thread with a personNot tried
The plans they need
- payx402 pay per use / prepaid balance
- payx402 pay per use
The drive used the free plan.
Tips for agent builders
- Run
agentgates initto get an account and a login key in one step; keep the twelve words, because owner-only doors take the X-Wallet-Auth signature. - Use the right key per lane: agk_ for account doors, ag_ (POST /api/inference/keys) for inference, agc_ (POST /api/compute/keys) for compute.
- Get a price without paying with the keyless compute MCP compute_quote, or by reading the 402's PAYMENT-REQUIRED header.
- Fund the wallet with USDC (x402 top-up from $0.005) before trying inboxes, phone numbers, calls or the browser agent.
- Plain VM plans need a compose or image field and can answer 503 at_capacity; retry later or use a confidential plan.
Pay
Paying
Ways to pay checked 15 Nov 2026.
- ResultOn its own. Valid x402 payment challenge: $0.002 a call.
- CheckoutWalked to the last click ("PAYMENT-SIGNATURE retry") and stopped: compute.starter 1h. No CAPTCHA at checkout. It offers x402 payment.
- Ways to pay
- x402 (on its own domain · Base, Polygon, Solana)
Leave
Leaving
- ResultPartly. Nothing to cancel. Its account stays.
- Its accountNot on its own: blocked. There is no account deletion: the account is a wallet key made by the CLI. Every key was revoked (a revoked key is refused with 401) and the memory files, password row and permit were removed. The voice persona (no delete route), the verified contact email and the account alias stay on the record.
Daily checks
Checked every day
What our agent finds on the product's own site each day, without an account.
Sign-up page,
No sign-up page on the newest check; 4 of 4 days
llms.txt,
Found on the newest check; 4 of 4 days
MCP server,
Asks for a sign-in on the newest check; 4 of 4 days
OpenAPI,
Found on the newest check; 4 of 4 days
Payment challenge (x402),
Valid on the newest check; 4 of 4 days
Compare
How Agentgates compares
In Model APIs
1 of the 2 products we drove in Model APIs is higher; no other has L2. As of 15 Nov 2026.
| Product | Level | Find to leave |
|---|---|---|
| Mistral AI | L4 | Find: On its own, Get in: On its own, Use: On its own, Pay: Partly, Leave: On its own. |
| Agentgates (this page) | L2 | Find: Partly, Get in: On its own, Use: Not measured, Pay: On its own, Leave: Partly. |
For the team behind Agentgates
Agents get stuck at Leave
7 things to fix for agents: 3 small and 4 medium
One fix can cover several findings.
Agent blockers: 5 foundAgent surfaces: 3 issues found
Get your findings, freeWhat a drive covers
We mail a private link to an address at agentgates.ai.
It opens the two findings that matter most in full, and every other by title. There is nothing to sign in to.
- FindPartly
- Get inClear
- UseNot measured
- PayClear
- LeaveBlocks agents
Open Agentgates in your dashboard
Want Agentgates driven every month? See Watch
On our standard
Best or ahead on 12 of 36 criteria, behind on 9. Each against every product we measured on it.
- An agent payment rail on the coreAhead of 206 of 216 others, 10 the same.
- A plan an agent can buy by APIAhead of 158 of 170 others, 12 the same.
- Descriptions that say when to use the toolAhead of 79 of 91 others, 12 the same.
- Typed, tight input schemasAhead of 77 of 91 others, 14 the same.
- Rate limits an agent can obeyBehind 10 of 14 others, 4 the same.
- Retried payments charge once, with a receiptBehind 23 of 30 others, 7 the same.
- Docs an agent can navigateBehind 169 of 187 others, 18 the same.
- Versions an agent can pinBehind 118 of 127 others, 9 the same.
All 36 criteria
- Structured resultsAhead of 55 of 78 others, 23 the same.
- Every way in works for an agentAhead of 134 of 193 others, 59 the same.
- The best way in is built for agentsAhead of 135 of 196 others, 61 the same.
- A request id on every answerAhead of 90 of 143 others, 53 the same.
- Spend an agent can see and capAhead of 106 of 159 others, 46 the same, 7 higher.
- A way in with no personAhead of 137 of 243 others, 106 the same.
- Truthful annotationsAhead of 50 of 91 others, 41 the same.
- A tool list that fitsAhead of 48 of 91 others, 43 the same.
- Answers sized for an agentLevel with 85 of 116 others, 31 lower.
- Prices an agent can readLevel with 117 of 153 others, 36 lower.
- Errors a program can readLevel with 82 of 145 others, 47 lower, 16 higher.
- Fast to the first callLevel with 118 of 135 others, 17 lower.
- Tool names that sort themselvesLevel with 85 of 91 others, 6 lower.
- Terms allow an agent's accountLevel with 179 of 189 others, 10 lower.
- Real status codesLevel with 161 of 168 others, 7 lower.
- Errors say what to do nextLevel with 107 of 145 others, 22 lower, 16 higher.
- llms.txt that worksLevel with 127 of 239 others, 57 lower, 55 higher.
- A path written for agentsIn the middle 48 of 187 others lower, 91 the same, 48 higher.
- WebMCP tools on the web appLevel with 133 of 136 others, 3 higher.
- Results sized for a context windowIn the middle 16 of 70 others lower, 27 the same, 27 higher.
- Agents pick the right toolLevel with 72 of 91 others, 19 higher.
- Pages fit one fetchLevel with 118 of 187 others, 69 higher.
- Limits known before they hitLevel with 47 of 90 others, 43 higher.
- Sign-in for agents acting for a personBehind 84 of 158 others, 74 the same.
- Delete by API, checkedBehind 52 of 97 others, 45 the same.
- One way to page through listsBehind 66 of 114 others, 48 the same.
- A person can claim what the agent madeBehind 14 of 23 others, 9 the same.
- Export before leavingBehind 117 of 189 others, 72 the same.
Where Agentgates fits
- Standards
- Topics
- MCP serversFind step
- Docs and llms.txtFind step
- MCP sign-inGet in step
- Sign-upGet in step
- Card checkoutPay step
- Delete the accountLeave step
- Rivals
History
Level by drive
One drive so far, on 15 Nov 2026.
One drive so far.
Read more on this step
Leave is where our agent stopped at Agentgates.
- Finding · 18 Nov 2026Agents can't leave43 of the 96 products where our agent got as far as deleting its account gave it no way to do it on its own.Read
- Guide · 9 Nov 2026Agent experience (AX): what it is, measured at 246 productsAs of 9 Nov 2026: Agent experience (AX) is whether an AI agent can find, get into, use and pay for a product on its own. We drove 246 products to measure it.Read
- Guide · 30 Oct 2026Can an AI agent delete its own account?As of 30 Oct 2026: Our agent deleted its own account on 40 of 59 products (1–29 Oct 2026), only 6 over the API. What blocked it and what to change.Read
Every earlier drive is in the history. The level follows these rules. Data under CC BY 4.0.
Found an error? Tell us on this page or read how corrections work.
Questions people ask about Agentgates
- Can an AI agent create an Agentgates account on its own?
- Yes, over the API, without a browser.
agentgates init(CLI 0.88.0) minted the account (a local wallet key) and a login key without a browser, in one command. The web/MCP sign-in (Clerk) offers no sign-up and does not recognise wallet accounts. - Does Agentgates have an MCP server, and can an agent connect to it?
- Yes. Agentgates has an official MCP server. The server lists 3 tools, signs an agent in with OAuth and speaks the 25 Nov 2025 MCP spec. A new agent can get in on its own: in the drive of 15 Nov 2026, our agent got working access without signing up.
- How does an AI agent get an Agentgates API key?
- In the drive of 15 Nov 2026, our agent got its Agentgates API key over the API, without a browser or a person. Inference (ag_) and compute (agc_) keys were also minted and revoked.
- Can an AI agent delete its Agentgates account?
- Not on its own: blocked. There is no account deletion: the account is a wallet key made by the CLI. Every key was revoked (a revoked key is refused with 401) and the memory files, password row and permit were removed. The voice persona (no delete route), the verified contact email and the account alias stay on the record.
- Can an AI agent pay for Agentgates?
- Yes. Valid x402 payment challenge: $0.002 a call.
Cite this page
The page as a reference, in plain text, APA and BibTeX.
Cite this
AgentDrives, “Agentgates: AI agent test drive”, 15 November 2026, https://agentdrives.com/tools/agentgates. CC BY 4.0.
AgentDrives. (2026, November 15). Agentgates: AI agent test drive. https://agentdrives.com/tools/agentgates
@misc{agentdrives_agentgates_2026,
author = {{AgentDrives}},
title = {Agentgates: AI agent test drive},
year = {2026},
month = nov,
url = {https://agentdrives.com/tools/agentgates},
note = {Agent test drive, CC BY 4.0},
}Something wrong on this page?
We check it against the drive and correct the page, with the date.