Autonomous bug-bounty framework for Claude Code and other AI coding tools with MCP servers for bounty platforms and writeup search
Server has 10 tools with complete input schemas and descriptions. Naming follows verb_noun convention (list_, get_, search_, submit_, save_). Descriptions are present and contextual (avg ~120 chars), meeting baseline minimums. However, output schemas are not documented in the source code provided, LLMs cannot see what fields to expect from responses. Parameter descriptions are adequate but lack format constraints (e.g., no enum for severity levels, no regex for platform IDs). Error handling is not visible in the provided code. Security concern: credentials passed via environment variables (good), but no evidence of input validation or injection protection in tool implementations.
Fetch the policy, testing restrictions, and safe harbor status for a bug bounty program.
Fetch the in-scope and out-of-scope assets for a bug bounty program. Returns structured scope data including asset types, eligibility, and bounty ranges. Use this BEFORE testing any target to populate scope files.
Get full writeup content by ID from the writeup database or local files.
List all supported bug bounty platforms and their configuration status.
Save a vulnerability report as a draft without submitting it. Useful for review before final submission.
Search disclosed reports and hacktivity for a bug bounty program to check for duplicates and understand past vulnerabilities.
Output schemas not documented. LLMs cannot see what fields to expect from tool responses (e.g., what does get_program_scope return? Is it a list of assets, a dict with 'in_scope'/'out_of_scope' keys?). This forces agents to guess and wastes tokens on trial-and-error parsing.
Severity parameter in submit_report and save_draft accepts free-form strings ('critical, high, medium, low, informational') but is not declared as an enum. LLMs may hallucinate invalid severity values. Should use enum constraint.
Platform parameter across all tools accepts free-form strings (hackerone, bugcrowd, intigriti, immunefi, yeswehack, etc.) but is not declared as an enum. Should enumerate valid platform IDs to prevent typos and hallucinations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Search for payloads by vulnerability type in rules/payloads.md. Returns structured payload packs with examples, variations, and detection bypasses.
Search for specific technique/vulnerability class patterns in the writeup corpus and local skills.
Semantic search (FAISS) or keyword search (fallback) across a bring-your-own bug-bounty / CTF writeup corpus. Agents query this for techniques, payloads, and prior art during hunting and validation.
Submit a vulnerability report to a bug bounty program. Handles platform-specific API calls and attachment uploads.
No error handling guidance visible in source. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., if submit_report fails due to invalid program, should the agent retry, call search_hacktivity, or ask the user?
Destructive tools (submit_report, save_draft) lack confirmation or dry-run capability. Agents can submit reports without review, risking duplicate or malformed submissions. Should support a dry-run mode or require explicit confirmation.