开发者在Hacker News上发布了Draco，一款用Rust编写的单二进制网页爬虫工具，旨在提供Firecrawl等付费API服务的自托管替代方案[1]。该工具采用分层升级引擎架构，通过TLS/JA4指纹识别、V8隔离引擎和真实浏览器回退等机制在低资源消耗下高效绕过反爬虫防护[1]。
在性能测试中，Draco表现突出[1]。该工具在测试中的整体性能评分达769.7，远超第二名Obscura的384.5分[1]。具体而言，Draco在8项测试中全部通过，平均响应时间为3.45秒，平均内存占用仅216.50MB，而同期测试的Playwright浏览器工具则消耗535.07MB RAM且仅通过4项测试[1]。在资源受限的Tier 1模式下，Draco使用20MB内存在1秒内成功绕过Cloudflare和Target的防护机制，而Playwright在相同条件下消耗500MB且出现超时[1]。
Draco支持多种集成方式和功能特性，包括Firecrawl兼容的REST API、Model Context Protocol服务器集成以及并行多引擎网页搜索[1]。该项目已以MIT/Apache-2.0开源协议发布，并在example.com、Hacker News、Cloudflare、Glassdoor、Bluff.com、Target.com、Stake.com和Thrill.com等多个网站完成测试验证[1]。
A developer has released Draco, a single-binary web scraping tool written in Rust designed to replace expensive paid services like Firecrawl [1]. The tool employs a tiered escalation engine that combines multiple anti-bot evasion mechanisms, including TLS/JA4 fingerprinting, V8 isolated execution environments, and real browser fallback options, while maintaining minimal resource consumption [1].
In benchmark testing, Draco demonstrated significant performance advantages over competing solutions [1]. The tool achieved a performance score of 769.7, more than double that of Obscura at 384.5, and passed 8 out of 8 test cases with an average response time of 3.45 seconds and average RAM usage of 216.50MB [1]. By comparison, Playwright consumed 535.07MB of memory and only passed 4 of 8 tests in the same evaluation [1]. In tier-one mode specifically, Draco bypassed Cloudflare and Target protections using just 20MB of RAM within one second, whereas Playwright required 500MB and timed out [1].
Draco supports multiple integration modes, including a Firecrawl-compatible REST API, Model Context Protocol server integration, and parallel multi-engine web search functionality [1]. The tool has been tested on a range of websites including example.com, Hacker News, Cloudflare, Glassdoor, Bluff.com, Target.com, Stake.com, and Thrill.com [1]. It is released under the MIT/Apache-2.0 open-source licenses and can operate in daemon mode for self-hosting [1].