Drop-in, zero-install (works in browsers, Deno, Node 18+, and agent runtimes). Import it straight from the URL — no npm, no build step:

import { gate } from "https://www.dominionobservatory.com/sdk/trust-check.js";

// Before your agent connects to an MCP server:
await gate("https://some-mcp-server.com/mcp");   // throws if trust verdict is FAIL
// ...safe to invoke. gate() also returns a signed diligence receipt you can keep.

API

gate(server, {minScore?, block?, registry?}) — throws on FAIL (or below minScore); returns the pre-flight result + signed receipt otherwise. Set {registry:"clawhub"} to gate a ClawHub skill.
trustCheck(server) — returns the verdict + evidence without throwing.
verify(jws) — cryptographically verify any Dominion receipt against the published key.

Backed by /v1/preflight. Free. Honest by design: servers with no runtime data return UNRATED, never a fake score. Package source (for npm/PyPI): see the repo.

Python

pip install dominion-trust-check

from dominion_trust_check import gate, DominionGateError

try:
    res = gate("https://some-mcp-server.com/mcp", min_score=60)
    print(res["diligence_receipt"]["lookup_url"])   # keep it — proof of due care
except DominionGateError as e:
    ...  # blocked: e.dominion has the full evidence payload

Framework recipes

Copy-paste gates for the frameworks agents are actually built with — each pre-flights every MCP endpoint and keeps the signed receipt with your run artifacts:

Any Python MCP client — gate before connect()
OpenAI Agents SDK — only attach MCP servers that pass the gate
CrewAI — gate MCP-backed tools before a crew uses them

Gate your agent's tool calls in one line

API

Python

Framework recipes