acrawl vs Firecrawl

Last reviewed October 2026.

Short answer: Firecrawl is primarily a hosted API for searching, scraping and interacting with websites, returning LLM-ready markdown and JSON at scale; its engine is open source (AGPL-3.0) and can be self-hosted. acrawl is a local, MIT-licensed agent you run yourself: you bring your own LLM key, it drives a browser on your machine or your real logged-in browser, and it adds DevTools observation and deterministic scripts.

At a glance

	acrawl	Firecrawl
Where it runs	Your machine, server or CI	Firecrawl’s cloud, or self-hosted
Pricing model	Free; you pay your own LLM provider	Hosted plans; self-hosting is free
Output	Structured JSON from a goal, or raw page content via tools	Markdown, JSON, screenshots, crawl results
Interaction	Click, fill, select, scroll, keys, device emulation	Actions and AI-driven Interact on scraped pages
Autonomous agent	Local agent with sub-agents, 25 LLM providers	Hosted /agent endpoint
Your logged-in browser	Yes, through the acrawl Bridge extension	Not its model; it uses its own browsers
DevTools observation	Network, console, WebSocket, performance, cookies, storage, coverage, axe-core	Not a focus
Repeatable runs without LLM calls	Deterministic JSON scripts	N/A
MCP server	Built in	Yes
License	MIT	AGPL-3.0 engine

Choose acrawl if

• You want everything to run locally with your own LLM key and no per-page fees.
• You need to work inside sites you’re logged into.
• You want to inspect how a page behaves, not just what it contains.

Choose Firecrawl if

• You want a managed API that handles crawling infrastructure and scale for you.
• You mostly need clean, LLM-ready content from many URLs.
• You’d rather not run browsers yourself.

The same job, two ways

Take “collect the name and price of every product in a 40-page catalog, every morning”.

With Firecrawl, you call its crawl or extract endpoint with the start URL and a schema, and it returns clean records from its own infrastructure. You pay per page on a hosted plan, or run the AGPL engine yourself.

With acrawl, you describe the job once:

acrawl > get the name and price of every product on example.com/catalog, following
         pagination, and save the crawl as a script called catalog

The agent works out the pagination and the selectors, returns the data, and saves the steps as a deterministic script. Every morning after that, a cron job runs it with no model at all:

acrawl run catalog --format jsonl --flatten --output catalog.jsonl

The first run costs some model tokens; the daily runs cost nothing but the machine they run on.

Using them together

They don’t compete for every job. A common split:

• Firecrawl for broad, shallow coverage: turning thousands of public URLs into Markdown for search or retrieval.
• acrawl for deep, interactive work: logging in with your own browser, working through multi-step flows, and debugging the pages your own team ships.

Both ship MCP servers, so a coding agent such as Claude Code can have both connected and pick the right one per task.

Common questions

Is acrawl a self-hosted Firecrawl? No. Firecrawl is a scraping service with an API; acrawl is an agent with a browser. acrawl has no HTTP API or job queue to host, and Firecrawl has no local agent that works through a task in your own browser.

Which is cheaper? For a few repeated jobs, acrawl, because scripts replay without model calls and there are no per-page fees. At very large scale, compare Firecrawl’s plan prices with your own LLM and server costs.

Does acrawl return Markdown like Firecrawl? Yes. navigate returns a page as markdown, fit_markdown (main content only), text or html, and the agent can shape results into whatever JSON you ask for.