acrawl
A single Rust binary that acts as an LLM-driven browser automation agent: it navigates, clicks and extracts data from a described goal in plain English, while also exposing the full DevTools surface as agent tools, plus a built-in MCP server.
acrawl is a browser automation agent that goes beyond simple navigate-and-click scraping: alongside standard actions it exposes network-request inspection with timing, deduplicated console-error analysis, page performance metrics such as time-to-first-byte and resource breakdown, axe-core accessibility audits, and network call interception or mocking, all as first-class tools an LLM agent can call. It ships as one self-contained Rust binary with no Python or Node runtime, downloading only a Chromium binary for the actual browser automation, so it can be dropped into a server or CI pipeline.
Static pages are fetched over plain HTTP for speed; acrawl detects JavaScript framework markers, auth redirects or unexpectedly short response bodies and escalates transparently to a full headless browser only when needed. Deterministic JSON scripts can define loops, conditionals and parallel branches that run with no LLM calls at all, falling back to the agent only when a page behaves unexpectedly, and child agents can fork onto separate browser tabs with independent state, coordinated by a URL-claiming registry that stops siblings from crawling the same page twice.
It supports 25 LLM providers and includes a built-in MCP server.