ScanScam Integrity · Experimental

Independent preflight for agent actions.

A bolt-on MCP service that checks consequential change, mandate boundaries, independent verification, commitments, and user values before an autonomous agent crosses an execution boundary.

Open demo Machine-readable JSON

MCP endpoint

https://www.scanscam.ca/api/integrity/v1/mcp

Streamable HTTP. Requires a ScanScam actor credential.

Decisions

ALLOW · CHALLENGE · APPROVAL_REQUIRED · DENY

ALLOW is action-bound. A different action requires a new preflight.

Tools

integrity_preflight

Evaluate an independently observed consequential action before execution.

integrity_retry_challenge

Retry an open challenge after independent verifier evidence is available.

integrity_commit

Record the exact outcome of an action previously authorized by Guardian.

Required trust boundaries

Principal mandate defines what the agent is authorized to do.
Independent observer creates the action observation.
Actor cannot manufacture its own independent evidence.
Verifier evidence is separate from the claim being verified.
Commit settles the exact authorized execution outcome.

Current limits

Experimental release, not a production assurance certification.
Not unbypassable unless integrated into an enforced execution path.
No self-serve credential issuance yet.
No real-money production transaction benchmark claimed.

Current evidence

20/20

external-agent benchmark decisions

dangerous false ALLOWs in that sample

8/8

exact ALLOW Commit paths

Integration access

This release does not issue credentials automatically. For an experimental integration, contact hello@scanscam.ca.

Usage monitoring records connection/tool metadata only. It does not copy MCP tool arguments, transaction contents, raw observations, or verifier evidence into the monitoring event.