mrbaeksang/deepcloak

DeepCloak is a research tool that reads web pages for you and writes a report with citations, even when those pages are protected by bot detection systems like Cloudflare, Datadome, Turnstile, or reCAPTCHA. Many research tools simply get blocked by these systems and quietly leave those pages out of their answer, without ever telling the user what was missed. The tool works by trying a normal, fast page fetch first. If that hits a bot wall, it escalates to what the project calls a stealth fetch, using a companion project called CloakBrowser to get past the block, and continues the research loop from another project called local deep research. At the end of every report, it lists exactly how many of these walls it had to bypass, so the user knows how much of the answer came from harder to reach sources. It can be used three ways: as a command line tool, as an MCP server that an AI agent like Claude can call directly, or as a Claude skill. Everything runs locally, meaning searches and page content stay on the user's own machine rather than being sent to a third party service, and no API key is required if a local model like Qwen is used instead of a cloud provider such as OpenAI or Anthropic. Setup is done through a Python package installer, followed by a one time setup command that downloads the stealth browser component. A person then needs an API key for a cloud AI provider, or a locally running model through Ollama, to generate the actual report text. The README is direct about the tradeoffs involved: the tool ignores site robots.txt rules by default, though this can be turned back on with a flag, and using it to bypass bot protection means the user is responsible for making sure they are actually allowed to access whatever they are reading. It is released under the MIT license.