Every demo runs against Clampd's real detection engine via gRPC. No simulations. No mocks. Real rules, real scores, real latency.
Attack payloads from SecLists, PayloadBox, OWASP, and Garak. Split detection rates for tool call security vs prompt scanning. Upload your own payloads. Share results.
LIVE detection rules 20 languages attack payloadsAgent-to-agent delegation validation. Tool descriptor verification catches rug-pulls. Delegation depth limits, cycle detection, and workflow enforcement.
LIVE Rug-pull detection Chain governanceAgent with db:read scope tries DROP TABLE - blocked. Least-privilege enforcement via Cedar policy engine. See exactly which scopes are required vs granted.
LIVE Cedar policies Scope exemptionsTerminate a rogue agent in under 25ms. Kill cascade walks the delegation tree - quarantining contacts and containing the blast radius automatically.
LIVE Cascade quarantine <25msScope tokens are single-use. Watch a token get minted, verified, then rejected on replay. Tampered tokens caught by signature verification.
LIVE ag-token 5 stepsConnect Claude.ai directly to Clampd-guarded MCP servers. Test attacks in natural language, see blocks in real time. Multi-agent scope isolation.
LIVE Claude.ai MCP serversNo signup. No API key. 3 free runs per hour.