# acrawl (AgenticCrawler)

> acrawl is an open-source browser agent and MCP server in one Rust binary. Describe a web task in plain English and it navigates, clicks, fills forms, watches the network and returns structured data.

Key facts: MIT-licensed, written in Rust. The agent has 43 tools; the MCP server (`acrawl mcp`, stdio) exposes 40: 32 browser and DevTools tools, 7 script tools and `run_goal`. It supports 25 LLM providers plus any OpenAI-compatible endpoint, installs into 17 MCP clients with `acrawl mcp install`, and replays saved JSON scripts with no LLM via `acrawl run`. Browser automation uses a bundled stealth browser (CloakBrowser, needs Node.js 20+) or the user's own browser through the acrawl Bridge extension. Source: https://github.com/Mingye-Lu/AgenticCrawler

## Docs

- [What is acrawl?](https://mingye-lu.github.io/AgenticCrawler/docs/): acrawl is an open-source browser agent and MCP server in one Rust binary: describe a web task in plain English and get structured data back.
- [Install acrawl](https://mingye-lu.github.io/AgenticCrawler/docs/install/): Install acrawl on Linux, macOS or Windows with one command. The script verifies the release checksum and sets up the stealth browser.
- [Quick start](https://mingye-lu.github.io/AgenticCrawler/docs/quickstart/): Connect an LLM provider and run your first acrawl crawl in a few minutes, from the terminal UI or a one-shot prompt.
- [Use acrawl as an MCP server](https://mingye-lu.github.io/AgenticCrawler/docs/mcp-server/): Add acrawl's MCP server to Claude Code, Cursor, VS Code, Zed and 13 other clients: 40 browser, DevTools and script tools over stdio.
- [LLM providers](https://mingye-lu.github.io/AgenticCrawler/docs/providers/): acrawl works with 25 LLM providers, including Anthropic, OpenAI, Gemini, Bedrock and Azure, plus any OpenAI-compatible endpoint such as Ollama.
- [How acrawl works](https://mingye-lu.github.io/AgenticCrawler/docs/how-it-works/): Inside acrawl: the agent loop, HTTP-first fetching with browser escalation, sub-agents with URL claiming, the browser backends, and known limitations.
- [Tools](https://mingye-lu.github.io/AgenticCrawler/docs/tools/): All 43 acrawl agent tools by group: navigation, interaction, extraction, DevTools observation, sub-agents and deterministic scripts.
- [Command-line reference](https://mingye-lu.github.io/AgenticCrawler/docs/cli/): Every acrawl command, flag, slash command, keybinding and exit code, for the terminal UI, one-shot prompts, saved scripts, CI and coding agents.
- [Configuration](https://mingye-lu.github.io/AgenticCrawler/docs/configuration/): Every acrawl setting: where the config files live, the defaults in settings.json, script limits, compaction, and the optional token and cost optimizations.
- [FAQ](https://mingye-lu.github.io/AgenticCrawler/docs/faq/): Answers about acrawl: what it is, supported LLMs, MCP clients, JavaScript sites, logins and CAPTCHAs, costs, and how it compares.

## Guides

- [Scrape JavaScript-rendered sites without code](https://mingye-lu.github.io/AgenticCrawler/docs/guides/scrape-js-sites/): Scrape JavaScript-rendered sites and SPAs from a plain-English goal: acrawl fetches over HTTP and escalates to a stealth browser when needed.
- [Repeatable crawls with zero LLM calls](https://mingye-lu.github.io/AgenticCrawler/docs/guides/deterministic-scripts/): Let the agent work out a crawl once, then replay it as a JSON script with loops and branches and zero LLM calls, from MCP or acrawl run.
- [Debug a live page: network, console, performance, accessibility](https://mingye-lu.github.io/AgenticCrawler/docs/guides/devtools-debugging/): Give an AI agent DevTools: network requests, console errors, WebSockets, performance, cookies, coverage and axe-core accessibility audits.
- [Use your real, logged-in browser](https://mingye-lu.github.io/AgenticCrawler/docs/guides/real-browser/): Drive your own logged-in Chrome, Edge or Brave with the acrawl Bridge extension, for sites behind a login or strict bot detection.

## Comparisons

- [acrawl vs other browser agents](https://mingye-lu.github.io/AgenticCrawler/docs/comparisons/overview/): How acrawl compares with browser-use, Playwright MCP and Firecrawl: reasoning loop, self-hosting, DevTools tools, scripts and license.
- [acrawl vs browser-use](https://mingye-lu.github.io/AgenticCrawler/docs/comparisons/vs-browser-use/): acrawl vs browser-use: two MIT-licensed browser agents compared on install, LLM choice, DevTools observation and deterministic scripts.
- [acrawl vs Playwright MCP](https://mingye-lu.github.io/AgenticCrawler/docs/comparisons/vs-playwright-mcp/): acrawl vs Playwright MCP: an MCP server with its own agent loop and DevTools tools compared with step-by-step browser control.
- [acrawl vs Firecrawl](https://mingye-lu.github.io/AgenticCrawler/docs/comparisons/vs-firecrawl/): acrawl vs Firecrawl: a self-hosted browser agent compared with a hosted scraping API, on control, cost, observation and license.

## Project

- [About acrawl](https://mingye-lu.github.io/AgenticCrawler/about/): acrawl is an MIT-licensed browser agent built in the open by Mingye Lu since March 2026: why it exists, how it is built, and who is behind it.

## Optional

- [Full docs as one text file](https://mingye-lu.github.io/AgenticCrawler/llms-full.txt)
- [README](https://github.com/Mingye-Lu/AgenticCrawler/blob/main/README.md)
- [Changelog](https://github.com/Mingye-Lu/AgenticCrawler/blob/main/CHANGELOG.md)