MCP

Use Agentic Bug Hunter from AI agents (Cursor, Claude Code, Codex, OpenCode). MCP adapts the existing research engine; it is not a separate scanner.

AI Agent → Agentic-Bug-Hunter MCP → Scope → Recon → Hunt → Validate → Report → Memory

Install

pip install 'mcp>=1.28'
./install.sh --agent mcp
bughunter mcp doctor
bughunter mcp serve

List tools at runtime:

bughunter mcp tools

Client config

Repo snippets:

Claude Code: mcp/bughunter-mcp/claude-config.json
OpenCode: mcp/bughunter-mcp/opencode-config.json

Example (Claude — merge into ~/.claude/settings.json):

{
  "mcpServers": {
    "bughunter": {
      "command": "python3",
      "args": ["mcp/bughunter-mcp/server.py"],
      "env": {
        "BBHUNT_MCP_APPROVE": "0"
      }
    }
  }
}

Run serve from a checkout where paths resolve, or point command/args at your install.

Scope and approval

Rule	Behavior
Scope first	Active tools blocked without allowlist domains
Approval	Active testing needs `approve=true` or `BBHUNT_MCP_APPROVE=1`
Discovered ≠ authorized	Recon hosts are not auto-promoted
No auto-submit	Report tools locate drafts only
Local-first	Research data stays local unless you configure external APIs

Policy path:

MCP tool → PolicyEngine.authorize_tool → ScopeChecker → AutopilotGuard → adapters → engines

Error codes

SCOPE_REQUIRED, OUT_OF_SCOPE, TARGET_REQUIRED, ACTIVE_TEST_APPROVAL_REQUIRED, REPORT_APPROVAL_REQUIRED, AUTHORIZATION_REQUIRED, INSUFFICIENT_EVIDENCE, RESEARCH_FAILED, CANCELLED

Tool catalog (25)

Tool	Notes
`bughunter_research`	Active — scope + approval
`bughunter_scope_check`	Read-only
`bughunter_get_scope`	Read-only
`bughunter_recon`	Active — scope + approval
`bughunter_attack_surface`	Read-only
`bughunter_subdomains`	Read-only
`bughunter_endpoints`	Read-only
`bughunter_parameters`	Read-only
`bughunter_hunt`	Active — scope + approval
`bughunter_hunt_class`	Active — scope + approval
`bughunter_list_findings`	Read-only
`bughunter_get_finding`	Read-only
`bughunter_validate`	Gated — scope + approval
`bughunter_get_validation`	Read-only
`bughunter_generate_report`	Read-only (draft)
`bughunter_get_report`	Read-only
`bughunter_get_evidence`	Read-only
`bughunter_memory_search`	Read-only (target filter)
`bughunter_memory_get`	Read-only (target filter)
`bughunter_memory_patterns`	Read-only (target filter)
`bughunter_leads`	Local lead board
`bughunter_next_lead`	Local lead board
`bughunter_update_lead`	Local lead board
`bughunter_program`	External read (program metadata)
`bughunter_program_policy`	External read (program policy)

Backed by existing tools/recon_engine.sh, tools/hunt.py, tools/validate.py, tools/lead_board.py, memory/*, and public HackerOne helpers under mcp/hackerone-mcp/.

Sibling MCP configs

Burp and Caido configs under mcp/ are third-party proxy integrations — see Integrations. Label experimental where your client setup is nonstandard.

Get started

Research

Integrate

Help

Install

Client config

Scope and approval

Error codes

Tool catalog (25)

Sibling MCP configs

​Install

​Client config

​Scope and approval

​Error codes

​Tool catalog (25)

​Sibling MCP configs

​Related

Install

Client config

Scope and approval

Error codes

Tool catalog (25)

Sibling MCP configs

Related