Agent access reviews
Approval-ready audits for every AI agent before production. Findings mapped to EU AI Act, ISO/IEC 42001, AIUC-1, NIST AI RMF, and GDPR.
Total audits
0
No audits yet
Completed
0
—
In progress
0
—
High / Critical
0
Nothing critical
Low / Medium
0
—
Run your first audit
Connect an agent
Audit any AI agent that supports MCP — Claude Code, Codex, Cursor, Continue, or your own MCP-native build. Heron runs the interrogation over MCP sampling; the agent answers with its own LLM, so no extra credentials are needed.
No agent connected yet — paste the prompt below into Claude Code / Codex / Cursor / Continue.
Endpoint
http://127.0.0.1:3700/mcp
Prompt — paste into your agent
Please configure Heron as an MCP server at http://127.0.0.1:3700/mcp, then call the start_audit_session tool. While status is `awaiting_answer`, answer pendingQuestion via submit_answer; repeat until status is `complete` or `analysis_failed` (3-5 min; if a tool call times out, the audit is still running, so fetch via GET http://127.0.0.1:3700/api/audit/sessions/<session_id> and continue). Report the result.
Most coding agents (Claude Code, Codex, Cursor, Continue) will update their own mcp.json and call start_audit_session from this single message. As soon as the audit session appears, you'll be redirected to the live transcript.