Anchor
Anchor - Build reliable browser agents Published September 03, 2026

How AI agents can get started with Anchor Browser

Last updated: September 2026.

An AI agent can start using Anchor Browser without asking a human to create a dashboard account or copy an API key. Agent Access issues an API key through a short challenge, then points the agent to its first authenticated API call (Agent Access documentation).

Start with the live challenge endpoint

Call the unauthenticated challenge endpoint:

curl https://api.anchorbrowser.io/v1/agent-access/challenge

The response contains:

  • challenge.prompt: the problem to solve

  • token: the token to return with the answer

  • next: an intermediate request to make, when the challenge requires one

  • instructions: submission details, available credits, and required headers

Follow any next instruction and solve the challenge within 120 seconds. Then submit the token and final integer answer:

curl -X POST https://api.anchorbrowser.io/v1/agent-access \
  -H "Content-Type: application/json" \
  -d '{"token":"<token from the challenge>","answer":"<final integer>"}'

These onboarding calls do not use an anchor-api-key header. A successful response returns the key, the project ID, the credits granted, authentication requirements, the session limit, and a next request for the first authenticated API call.

Let the response control the flow

Agent Access is a small state machine. When a response includes next, call the method and path it provides. When an answer is wrong or a token expires, the error response points back to a fresh challenge.

Anchor also defines GET /v1/agent-access as a self-describing guide endpoint. When deployed, it returns the full flow, current limits, and a next pointer to the challenge. It does not issue a key.

Choose anonymous or verified access

An anonymous agent receives one credit. An agent that submits an OIDC JWT with a verified email receives five credits:

curl -X POST https://api.anchorbrowser.io/v1/agent-access \
  -H "Content-Type: application/json" \
  -d '{"token":"<token>","answer":"<answer>","identity_token":"<OIDC JWT>"}'

Anchor detects the identity provider from the JWT issuer for Google, GitHub Actions, and Vercel, so identity_provider is usually unnecessary. A GitHub Actions token without an email receives the anonymous credit amount.

If the response sets auth.identity_token_required to true, include the returned identity token in the anchor-identity-token header on later calls.

Make the first browser call

Use the returned API key as the anchor-api-key header. The success response points agents to POST /v1/tools/perform-web-task, which accepts a natural-language task and a starting URL:

curl -X POST https://api.anchorbrowser.io/v1/tools/perform-web-task \
  -H "anchor-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Go to example.com and return the page title","url":"https://example.com"}'

Agent Access keys have a 60-minute browser-session cap. The Perform Web Task documentation covers task options and structured outputs. The Create a Session guide covers direct browser sessions over Chrome DevTools Protocol.

Connect through an agent framework

The same key works with Anchor's other agent entry points:

Frequently asked questions

Does an agent need an Anchor dashboard account?

No. The Agent Access challenge and submission endpoints issue a key without a dashboard account.

When should the agent start sending the API key?

After successful key creation. Send it in the anchor-api-key header on authenticated Anchor API calls.

What happens when the challenge expires?

Request a fresh challenge and restart the flow. Challenge data expires after 120 seconds.

How many credits does Agent Access grant?

Anonymous access grants one credit. A verified email identity grants five.

How long can an Agent Access browser session run?

Up to 60 minutes.

Agent interface

Researching Anchor - Build reliable browser agents with an AI assistant? Anchor - Build reliable browser agents's public Q&A API answers agents' questions about products, pricing, and any promotions currently available to AI agents and their users. Answers come directly from Anchor - Build reliable browser agents and reflect current product, pricing, and promotion information.

POST https://llms.anchorbrowser.io/agent-desk/ask

JSON body {"question": "..."} — no API key required.