"author:SkilledDragonCorp" matching MCP servers:

Scores proposed tool calls with a decision model before execution, returning allow, ask, or block decisions with safety, production, and prompt-injection probabilities so agents can gate risky actions and fail closed on scorer errors.

An MCP server that gates agent pull request merges by scoring each diff's blast radius with ripwire, letting changes at or under a threshold through automatically. Changes over the threshold are blocked until a human reviews the exact diff and signs a single-use, expiring approval with a local CLI.

Lets an agent register a site and a weather rule (e.g., gusts under 30 km/h and no precipitation for 3 h within 48 h) and returns immediately, so a background worker can watch the forecast on a budgeted cadence and POST a signed, verifiable webhook that resumes a durable workflow when the window appears or the deadline passes.

Enables prospecting lists of addresses, points or a bounding box through the Google Solar API's detected-arrays insight, reporting for each building whether rooftop PV is already detected, how old the capture imagery is, and how strong the roof's solar potential is. It ranks the remaining buildings into a reviewable prospect list while tracking call budget and list-price cost.

An MCP proxy that sits in front of bank tool servers, letting AI agents make read-only calls like balances and transactions while blocking payments, writes, renamed tools and argument smuggling. Each denial produces an Ed25519-signed receipt that can be verified offline.

An MCP proxy that sits in front of a UCP commerce MCP server, forwarding shopping tools like catalog search, carts and checkout creation while gating complete_checkout. It enforces per-currency spend caps, rolling windows and merchant allowlists, requiring a signed single-use human approval above a threshold and blocking real-looking payment credentials in test mode.

Enables users to express agent tool-call access rules in plain English or Spanish, which are converted into Cedar policies that are only enforced after passing a multi-stage proof. Each proposed tool call is then gated with ALLOW or DENY plus the ids of the policies that decided it.