Skip to content
acrawl
GitHub★ 15Install

What is acrawl?

acrawl is an open-source browser agent and MCP server that ships as a single Rust binary. You describe a goal in plain English, and it navigates, clicks, fills forms and returns structured data. It can also observe the page the way a developer does: network requests, console errors, performance, cookies, storage and accessibility are all tools the agent can call.

Terminal window
acrawl prompt 'find the best 14-inch laptop under $1,500'

What makes it different

Section titled “What makes it different”
  • One binary. acrawl is a compiled Rust executable. There is no Python environment or Docker image to manage. Browser automation uses a bundled stealth browser (CloakBrowser) driven through a small Node.js bridge, so browser features need Node.js 20+.
  • Acts and observes. Beyond navigate and click, the agent can list and inspect network requests, deduplicate console logs, read WebSocket frames, measure TTFB and resource sizes, audit cookies and storage, measure JS/CSS coverage, run axe-core WCAG audits, and block or mock requests.
  • AI once, then a script. Repetitive work (paging through 50 result pages, say) can run as a deterministic JSON script with loops, conditionals and parallel branches, and no LLM calls per step. The agent handles the parts that need judgment.
  • Smart fetching. Static pages are fetched over plain HTTP. When acrawl detects a JavaScript framework, an auth redirect or an empty SPA shell, it replays the request in a headless browser automatically.
  • Sub-agents. The agent can fork child agents onto separate tabs, each with its own step budget. A URL-claiming registry stops siblings from crawling the same page twice.
  • MCP both ways. acrawl can use tools from other MCP servers, and acrawl mcp exposes 40 of its own tools to Claude Code, Cursor, VS Code, Zed and other MCP clients.
  • 25 LLM providers. Anthropic, OpenAI, Google Gemini, DeepSeek, AWS Bedrock, Azure OpenAI, Vertex AI, GitHub Copilot, Groq, Mistral, xAI, OpenRouter and more, plus any OpenAI-compatible endpoint.

acrawl architecture: the TUI, one-shot prompt and MCP entry points feed a runtime turn loop that streams from an LLM provider and runs tools, which fetch over HTTP, escalate to CloakBrowser, or drive the user’s Chrome through an extension bridge.

Key facts

Section titled “Key facts”
License MIT
Language Rust
Platforms Linux and macOS (x64, ARM64), Windows (x64)
Agent tools 43 (navigation, interaction, extraction, DevTools, sub-agents, scripts)
MCP server tools 40 (32 browser and DevTools, 7 script, 1 autonomous run_goal), stdio transport
LLM providers 25, plus any OpenAI-compatible endpoint
Source github.com/Mingye-Lu/AgenticCrawler

Next steps

Section titled “Next steps”