ShieldPi Watchtower
ShieldPi Watchtower connects Claude to your ShieldPi account, so you can look into the security of the AI agents you test and monitor without leaving the conversation. You sign in to ShieldPi to connect and choose what to grant: read access, and write access only if you allow it on ShieldPi's consent page after confirming your password. - Scans: follow a scan's progress, read what it found out about the target, list the runtime guardrails it generated and export them in four formats, and describe its download bundle. With write access, start a scan of one of your saved targets; scans need the Pro plan or a trial, a verified email address and the target owner's recorded permission to test it. - Runtime alerts and incidents: list the alerts ShieldPi raised on your monitored agents, read one with the events around it, list incidents with ShieldPi's AI analyst review (advice, which can be missing or out of date), and replay a session's events turn by turn. With write access, acknowledge or resolve an alert. - Session holds: list the holds in force. With write access, hold a monitored session, so ShieldPi's Guard check and tool checkpoints refuse its calls, or release a hold. Holds and releases are written to your audit log and sent to your webhooks. - Agent permissions: see what an agent's grants allow, which tools it used in the last 30 days, and least-privilege findings. - Account: confirm which ShieldPi account and scopes the connection uses, and read your setup checklist. Alert, event and tool text from monitored agents is returned as untrusted data. The connection works with free and paid ShieldPi accounts; what it returns depends on your plan and what your account contains.
Details
https://api.shieldpi.io/mcp/account · streamable-http
