Ghostfox is an alpha open-source browser stack intended for AI agents that need to operate websites while keeping browser identities and session data under the operator’s control. The project combines Firefox, anti-detect patches derived from Camoufox, and a Ghostfox engine that places fingerprint spoofing in C++-level browser patches rather than injected JavaScript. A separate Rust runtime manages sessions, identities, MCP communication, evidence, and browser actions. The repository exposes 43 MCP tools covering session creation, accessibility inspection, reference-based clicks and typing, waiting, file uploads, screenshots, page debugging, network inspection, vision, OCR, identity generation, auditing, and action confirmation. It also includes local captcha tools for eight listed captcha families, including GeeTest, hCaptcha, ordinary OCR, rotation challenges, and Turnstile or TikTok recipes; the project says these flows were tested on production sites, including Bilibili, hCaptcha-protected signups, and TikTok OAuth with OTP. Native solvers are described as local, while optional vision integrations can use external models such as Cloudflare Workers AI, GLM, or Qwen. Ghostfox generates device profiles intended to keep platform, screen, GPU, fonts, timezone, locale, and related signals coherent, and reports 500/500 identity-audit results. Persistent profiles can retain cookies, storage, and identity data, and each session records append-only events, page snapshots, and identity information for later review. The project supports Python, npm, Docker, source builds, and the official MCP Registry, with prebuilt distribution described for Linux x86_64. Its engine is licensed under MPL-2.0, while the Rust runtime uses MIT or Apache-2.0. The stated status is v0.7 alpha, and the maintainers caution that it should only be used for authorized testing or research.
