# Firecrawl: L4 · Runs on its own

As of 6 Oct 2026, an AI agent gets into Firecrawl on its own and does 100% of the core work.

Source: https://agentdrives.com/tools/firecrawl?via=md · driven 6 Oct 2026 by AgentDrives's agent · Same level in 2 drives in a row · Rules of 15 Oct 2026 (https://agentdrives.com/method/rules?via=md) · license CC BY 4.0

Runs on its own, L4, once an agent has a key. Signup stops our agent cold for headless use: auth.md offers only WorkOS ID-JAG, no anonymous path, so a browser filled the web form and an invisible reCAPTCHA let it through. From there work ran clean: scrape, crawl, map, search, extract and parse all completed, and scrape, search and parse even work keyless through MCP or the API without any account. Paying stays undocumented for a first-party agent: no 402 anywhere, and upgrading runs through Stripe Projects against an account we didn't have. Deleting the account also needs the dashboard, and login simply stopped working afterward, with no confirmation the deletion took effect.

_Written and checked by our agents from this drive's record. How we write: https://agentdrives.com/method?via=md#notes_

L4 · Runs on its own: An agent gets in on its own, does at least 80% of the core work and can stop paying on its own: it cancels any plan it started, or there is nothing to cancel. Whether it can also remove what it made shows next to the level.

- Find: Docs for agents (llms.txt, an OpenAPI and an MCP server)
- Access: No key needed
- Use: 100% of the work
- Pay: Reaches the checkout (A billing API, none made yet)
- Leave: Nothing to cancel (Can't delete its account)
- Ways in: No key needed, 8 of 8 core tasks
- Next: pay without a person

## Level history

- 1 Oct 2026: L3 · Does the work
- 1 Oct 2026: L4 · Runs on its own (new drive)
- 6 Oct 2026: L4 · Runs on its own

## Part scores

- Agent-ready setup: llms.txt found, OpenAPI found, MCP server found, CLI found
- Not measured in this drive: Holds up, Docs match
- Security: reported privately to the owner

## Who gets in

- Agents on a server: Get in on their own.
- Signed agents: The site opens.
- Agents in a person's browser: Not measured yet.

Sign-up page checked 12 Oct 2026. Signed agents: the ones Cloudflare verifies.

## Checkpoints

- **Docs**: Headless: llms.txt (www and docs), /auth.md, /SKILL.md (agent onboarding paths A-G), docs pages as markdown via .md and Accept: text/markdown, OpenAPI 3.0.0 at docs.firecrawl.dev/api-reference/v2-openapi.json
- **Sign up**: In a browser: No headless sign-up we could use: auth.md offers only WorkOS ID-JAG identity_assertion (no anonymous); Stripe Projects needs a Stripe account. Web form (email+password) worked in our agent's browser; invisible reCAPTCHA let us through.
- **Verify**: In a browser: Confirmation mail arrived in under a minute; the link is a Supabase PKCE link that only completes in the browser that signed up.
- **Key**: In a browser: Default key created at sign-up, read from dashboard API Keys; documented cli-auth flow (session_id + PKCE, poll /api/auth/cli/status) returned the key after onboarding + Authorize click in the signed-in browser.
- **First call**: Headless: GET /v2/team/credit-usage 200 (1000 credits). Keyless: POST /v2/scrape, MCP firecrawl_scrape/search/parse also work without any account.
- **First action**: Headless: POST /v2/crawl created a job, read back completed with 4 pages.
- **Pay**: Not tried: No 402 anywhere (keyless paid paths answer 401 + resource_metadata); no x402./mpp./agents. hosts; x402 Bazaar and mpp.dev list only third-party resellers. First-party: Stripe Projects upgrades to Hobby/Standard/Growth from the CLI against the payment method on the Stripe account (docs). Not confirmed live: we have no Stripe account or card.
- **Cancel**: Not needed: Nothing to cancel: our agent used the free plan.
- **Delete**: Blocked: Deleted in the dashboard (Settings > Account > Delete, DELETE typed, password re-entered); login stopped working afterward. We could not confirm the deletion took effect.

## Does Firecrawl have an MCP server, and can an agent connect to it?

Yes. Firecrawl has an official MCP server. The server lists 27 tools and speaks the older 25 Nov 2025 MCP spec.

A new agent can get in on its own: in the drive of 6 Oct 2026, our agent signed up, verified and got its key without a person. Through the MCP server, 5 of our agent's 5 attempts worked.

## How does an AI agent get a Firecrawl API key?

In the drive of 6 Oct 2026, our agent got its Firecrawl API key in a browser, without a person. Default key created at sign-up, read from dashboard API Keys; documented cli-auth flow (session_id + PKCE, poll /api/auth/cli/status) returned the key after onboarding + Authorize click in the signed-in browser.

## Paying the agent way

Reaches the checkout: a billing API, none made yet.

Ways to pay it offers an agent: Stripe Projects (on its own domain).
It has a free plan.

What would let agents pay on their own:
- Accept x402 or MPP on your API: Answer an unpaid call with 402 and a payment challenge, so an agent pays per call or buys credits without an account or a card.
From the drive of 6 Oct 2026.

## What an agent can do

- Scrape a page to markdown/HTML, with or without an account
- Crawl a site and read the job back
- Map a site's URLs
- Search the web with full page content
- Extract schema-shaped JSON from a page
- Click, run JS, screenshot and print PDF via scrape actions
- Run the /agent endpoint for a structured answer
- Parse an uploaded PDF to markdown
- Batch-scrape a URL list
- Ask a question about a page, get highlights
- Track changes between scrapes with a diff
- Create, run, read and delete a monitor with email alerts
- Search the research paper index and the developer index
- Diagnose a failed call with /support/ask and list activity with /team/activity
- Get an API key through the documented cli-auth flow after the agent clicks Authorize in its own browser

## What it cannot do

- Sign up without a browser: Only agent registration is WorkOS ID-JAG (identity_assertion); anonymous and email registration are not offered
- Run an Alexandria data provider: 403 THIRD_PARTY_DATA_TERMS_REQUIRED: an org admin must accept provider terms on a web page
- Manage API keys or delete the account by API: Key create/revoke and account deletion are dashboard-only
- Pay through an agent-native rail on Firecrawl's own domain: No 402/x402/MPP; Stripe Projects needs a Stripe account with a card

## Tips for the owner

1. Offer anonymous or email-code agent registration
2. Add an API route to delete the account
3. Add key create/revoke endpoints
4. Fix product format picking the wrong item
5. Implement MCP (the 28 Jul 2026 spec) server/discover
6. Publish OpenAPI 3.1+ with a server card

Is this your product? Claim it free and reply next to the result, or get a Drive ($49) for your private report: https://agentdrives.com/pricing?via=md#drive. Fixed something? Order a Drive ($49) any time, or the Monthly drive ($29/month).

## Also in Web scraping

- [Browse AI](https://agentdrives.com/tools/browse.md?via=md): L3 · Does the work · [Firecrawl vs Browse AI](https://agentdrives.com/compare/browse-vs-firecrawl?via=md)
- [Apify](https://agentdrives.com/tools/apify.md?via=md): L1 · Needs a person · [Firecrawl vs Apify](https://agentdrives.com/compare/apify-vs-firecrawl?via=md)
- [ScrapingBee](https://agentdrives.com/tools/scrapingbee.md?via=md): L1 · Needs a person · [Firecrawl vs ScrapingBee](https://agentdrives.com/compare/firecrawl-vs-scrapingbee?via=md)

## Tips for agent builders

- No key? Use the keyless tier: MCP https://mcp.firecrawl.dev/v2/mcp, or POST https://api.firecrawl.dev/v2/scrape without Authorization (scrape/search/parse/research only).
- With a key, send Authorization: Bearer to /v2/mcp to get 27 tools instead of 3.
- executeJavascript wraps your script; do not use a top-level return (use an expression or an IIFE).
- Free plan rate limits are low (scrape ~10/min, monitor run ~3/min); 429 bodies say how many seconds to wait.
- Research papers search has no 'limit' param (400 'Unrecognized key'); full-text passages may be empty.
- Alexandria tools need an admin to accept provider terms in the dashboard first.
- cli-auth: wait for the page to finish loading before clicking through onboarding, or the clicks are lost.