What happened
Claude-BugHunter is a repository of skills and commands for vulnerability research and external red-team work. Its GitHub page describes a bundle of 82 skills, 15 slash commands and 681 patterns drawn from disclosed reports across 24 vulnerability classes. The repository also contains MCP integrations and harness setup tooling.
The latest repository change, dated November 18, 2026, wires the HackerOne MCP into multiple harnesses. The project says that the earlier HackerOne integration was installed only into Claude Code. The new setup script can configure HackerOne for OpenCode, Codex, Hermes and AntiGravity, in addition to Claude Code support. Its installer options include --hackerone-mcp and -HackeroneMcp; the setup script accepts --hackerone to select that server. The change is designed to coexist with the repository's Burp integration, so both servers can be configured in one run.
The HackerOne MCP exposes public tools for searching disclosed reports, retrieving program information and retrieving program scope. Search supports a program filter by team handle. When HACKERONE_API_USERNAME and HACKERONE_API_TOKEN are configured, it also exposes read-only tools for listing and retrieving the user's programs and reports. The plugin configuration treats the token as sensitive and uses Claude Code's secure storage; the server starts only while the plugin is enabled. For the other harnesses, the setup script passes the HackerOne environment variables only when they are set.
The integration uses a standard-I/O MCP server implemented in Python. The repository describes its server as using the Python standard library, and the multi-harness setup points at the server within the repository. The release also adds documentation and tests for source resolution and harness wiring. It is an integration and access change rather than a new vulnerability-finding model: its practical effect is to make HackerOne data available to more of the supported agent environments.
What the archive adds
New Runtime covered Anthropic's Claude Security continuous monitoring and fix tracking on October 22, 2026. That was a separate product update; Claude-BugHunter is a new event, extending a repository's security research workflow with HackerOne MCP access across harnesses.
Trend signal
This supports agent protocols become the interoperability layer: MCP provides a common way to expose HackerOne operations to several harnesses instead of implementing a separate integration for each agent environment. The signal is specific to a project integration, not evidence that MCP has become a universal or dominant security-tool interface.
Why now
The repository's November 18 change explicitly addresses the gap between its Claude Code integration and the other harnesses it already supported for Burp. The new wiring makes HackerOne access fit the project's existing multi-harness setup rather than leaving it attached to a single environment.
Connections to other events
New Runtime's earlier coverage described Claude Security's continuous code monitoring and fix tracking on October 22, 2026. Both concern security work with AI, but this repository update is distinct: it connects an agent harness to HackerOne program data and disclosed reports rather than monitoring code changes.
Why it matters
Security teams often work across several agent environments and rely on different sources for vulnerability research and program context. Making HackerOne queries available through MCP can reduce the need to switch tools or build a bespoke connector for every harness. Public search and scope retrieval can help an agent ground research in disclosed findings and program boundaries; authenticated read-only operations can expose a team's own program and report information when configured.
The important operational distinction is access. Public tools are usable without HackerOne credentials, while the private account operations are gated on a username and token. Teams should treat that token as a meaningful permission boundary, scope who can configure it, and validate the agent's use of program scope before allowing testing. MCP makes tool access more portable; it does not by itself establish that an agent's security conclusions are correct or that a target is authorized.
Limits and uncertainty
The repository describes the tool surface and harness setup, but this release does not establish how accurately agents use the results, how much time the integration saves, or whether teams have deployed it in production. The authenticated tools are read-only, and access depends on configuring HackerOne credentials. The repository's stated counts of skills, commands and report patterns describe the project contents, not independently measured security outcomes. Teams should validate the server's behavior and permissions in their own environment before connecting sensitive accounts.
Skill to build now
If you maintain an agent workflow for security research, try the integration first against public disclosed-report search and a test program. Confirm that the returned scope is visible in the agent's working context before using it to guide any testing. Add the authenticated read-only tools only when a workflow needs private program or report data, and keep the token in the harness's secret-storage mechanism rather than in prompts or repository files.
Reader takeaway
Claude-BugHunter is making HackerOne data a reusable MCP capability across several agent harnesses. The useful change is portability: security teams can connect program context and disclosed reports without treating Claude Code as the only supported environment. Credentials remain a separate access decision, and the integration should not be mistaken for validation of an agent's security findings.