h1-brain

What it is

h1-brain is a Python MCP server (MIT license, by Patrik Grobshäuser, repo PatrikFehrenbach/h1-brain) that lets Claude Desktop, Claude Code or any MCP-compatible client query a researcher's HackerOne history. It stores data in local SQLite files. The repo includes disclosed_reports.db, with 3,600+ public bounty-awarded reports that have real write-ups. Each entry has title, details, weakness type, program, asset, CVEs and bounty amount when available.

How it works

Two databases sit behind the tools. h1_data.db holds your own reports, programs, scopes and activity timelines. It is auto-created and gitignored. disclosed_reports.db is read-only community data. Personal tools (search_reports, get_report) and public tools (search_disclosed_reports, get_disclosed_report) are kept separate.

hack(handle) runs this sequence: fetch current scope from the HackerOne API, pull your reports for that program, analyze weakness patterns across all your programs, find bounty-eligible assets you have not touched, cross-reference public disclosures for the program, then build a briefing. Its "Suggested Attack Vectors" section lists weaknesses that paid on other programs but have not been found on this one. The briefing is loaded from hack_instructions.md and ends with instructions that put the AI in attack mode.

Preconditions / Affected

• Python 3.10+.
• A HackerOne API token and username.
• Your own program access: fetch_programs pulls only programs you can access.
• The personal database is empty until you sync it.

What to do

Install and register with Claude Code:

git clone https://github.com/PatrikFehrenbach/h1-brain.git
cd h1-brain
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
claude mcp add h1-brain \
  -e H1_USERNAME=your_hackerone_username \
  -e H1_API_TOKEN=your_api_token \
  -- /path/to/h1-brain/venv/bin/python /path/to/h1-brain/server.py

For Claude Desktop, add an mcpServers entry named h1-brain to ~/Library/Application Support/Claude/claude_desktop_config.json. Set command to the venv Python, args to server.py, and env to H1_USERNAME and H1_API_TOKEN. Restart Claude Desktop afterwards.

Then populate the personal database. fetch_rewarded_reports is described as the most important step. fetch_programs follows it. Re-run both periodically to sync new reports.

To include non-rewarded reports (informative, triaged, duplicate, new) and all comment threads for one program, call fetch_program_reports(handle). An optional out_dir exports each report as a markdown file.

Caveats

• Your API token sits in the MCP client config as plaintext environment variables, and report content goes to the connected AI.
• The page does not describe how the public dataset was collected or how often it is refreshed.
• The page does not say how the attack vector suggestions are scored. They compare weaknesses across programs.
• The tool is meant for programs where you are authorized to test, and the briefing pushes the AI into offensive mode. The page gives no scope or safety guardrails.