folk vs Twenty for AI agents

As of 6 Oct 2026, Twenty gets an AI agent further: L4 · Runs on its own against L1 · Needs a person.

folkL1Needs a person
  • Find: Docs for agents: llms.txt, an OpenAPI and an MCP server.
  • Access: Can't verify its account.
  • Use: Work not reached.
  • Pay: Payment not reached.
  • Leave: Leaving not reached.

Stopped at verify: blocked. Next: let an agent verify its account on its own

Gets in only with a person: L1 · Needs a person. Our agent started sign-up itself at app.folk.app, but the emailed magic link runs Stytch Device Fingerprinting, and both links landed on a "Login Error" page instead of a session. Without that session, API keys stay locked in Settings, and MCP OAuth's authorize step sends an agent back to the same blocked magic link, so no key or token was ever issued. No CRM work was done through API, MCP or CLI. Docs are thorough: llms.txt, Markdown pages, OpenAPI, MCP guides. Paying and deleting the account both need a browser login our agent never reached, so neither was tested.

Driven 6 Oct 2026 · The drive and its evidence
TwentyL4Runs on its own
  • Find: Docs for agents: llms.txt, an OpenAPI and an MCP server.
  • Access: Sign-up in a browser.
  • Use: 80% of the work.
  • Pay: Payment not reached.
  • Leave: Leaves on its own: Cancels on its own, deletes its account.

Got in; payment not reached. Next: pay without a person

Runs on its own: L4 · Runs on its own. Signup, onboarding and the first key need a browser, and invisible reCAPTCHA let our agent through with no challenge shown. Paying is where it stopped: a trial starts without a card, but clicking add credit card at checkout did nothing in our agent's browser, so a person has to add the card. Short of that, Twenty worked headless: REST and an official MCP server let it create a company and read it back, with tools covering records, custom objects, views and workflows. Cancelling and deleting the workspace both worked on their own, with no card ever added.

Gets furtherDriven 2 Oct 2026 · The drive and its evidence

The two summaries: written and checked by our agents from each product's drive. How we write.

Where each one stops.

folkStopped at verify: blocked

TwentyGot in; payment not reached

folk and Twenty, checkpoint by checkpoint
CheckpointfolkTwenty
DocsHeadlessHeadless
Sign upIn a browserIn a browser
VerifyBlockedIn a browser
KeyNot reachedIn a browser
First callNot reachedHeadless
First actionNot reachedHeadless
PayNot reachedNot reached
CancelNot reachedIn a browser
DeleteNot reachedIn a browser

Headless or in a browser.

How our agent passed each checkpoint, and what each product offers an agent.

Checkpoints by route, API, MCP, CLI and SDK
Checkpoints by how our agent passed them, and what each product offers an agent
folkTwenty
Headless (checkpoints)13
In a browser (checkpoints)15
Needs a human (checkpoints)00
Blocked (checkpoints)10
Not tested (checkpoints)61
Work doneNot measured80% of the core
APIYesYes
OpenAPIYesYes
MCP serverYes (official)Yes (official)
CLINoYes
SDKNoNo
llms.txtYesYes

Bot checks, and where a person was needed.

Walls our agent met
folkTwenty
Bot checkStytch Device Fingerprinting (magic-link redirect) stopped it at verify, without a visible challengereCAPTCHA (invisible) let it through at sign up
Wall typeA check of our network and browserNo bot-check wall
First wallStytch fingerprint check blocks magic-link loginGot in; payment not reached
A person helpedNoNo

A check of our network and browser: It judged where our agent connects from and its browser, not a person. An agent that works on its own from a data center doesn't get through unless the owner lets signed agents in. How agents in a person's own browser fare isn't measured yet. Where we drive from.

Level and part scores.

folk and Twenty: level and part scores, 0 to 100
folkTwenty
LevelL1 · Needs a personL4 · Runs on its own
Agent-ready setup90 of 100, good100 of 100, good
StandardsMCP client registration (current) · OpenAPI 3.1.0 (1 version behind)None recorded

Holds up: not measured in these drives. Docs match is measured in an Audit. What the scores mean.

Pay, cancel and delete.

Paying, cancelling and deleting the account
folkTwenty
PayPayment not reachedPayment not reached
Free plan or trialTrial, no cardTrial, no card
CancelNot reachedIn a browser
Delete the accountNot reachedIn a browser
How each one takes payment
  • folk: Payment not reached.
  • Twenty: Payment not reached.

folk vs Twenty: the short answers.

Which can an AI agent use, folk or Twenty?

Twenty gets an AI agent further: L4 · Runs on its own against L1 · Needs a person. With folk our agent was stopped at verify (blocked); with Twenty it got in (payment not reached). Driven 6 Oct 2026 (folk) and 2 Oct 2026 (Twenty).

Can an AI agent sign up for folk and Twenty on its own?

folk: yes, in a browser, but a later step stopped it. Twenty: yes, in a browser.

Do folk and Twenty have an MCP server?

folk: yes (official). Twenty: yes (official).

Can an AI agent delete its folk or Twenty account?

folk: not reached. Twenty: yes, in a browser.

The drives and the category.

Put your own shortlist side by side.

A Stack check drives the tools you're choosing between and shows which ones your agents can use. $490 for three tools, ready in 24 hours.