Paste an npm or PyPI package name — or a GitHub repo — and get a deterministic A–F Trust Score from its real source in seconds. No signup, no API key. Then share the result or grab a live README badge. The same open-source engine that powers the registry, the free API and the CLI.
The Capability-Flow Trust Model scores a server the way an attacker reasons about it. It is a named, versioned methodology built from scratch for MCP — not a wrapper around any existing scanner.
Two independent axes. A browser driver or code-runner is powerful (large blast radius) but still trustworthy — the grade never punishes a server for what it's built to do, only for signs of malice or negligence.
It maps every tool to the roles it can play and hunts the lethal trifecta — untrusted input → sensitive data → an exfiltration path co-reachable in one session. That's the shape behind real MCP data-exfil exploits.
Tool poisoning, hidden-Unicode payloads, embedded secrets, obfuscation, command-injection sinks and rug-pull drift — found in the actual source you'd install, each with its evidence. No LLM, no telemetry, no opinion.
Every rule is calibrated against a live corpus of 30,000+ scanned MCP servers, audited for both false positives and false negatives before each release — and it keeps improving.
Identical deterministic results everywhere — the same input always yields the same A–F grade.
Paste an npm or PyPI package name, or a GitHub repo. Nothing to install, no key. Great for a quick check or to share a link to a server's grade.
Scan a server ↑Runs offline and local-first: scan a manifest, a packed .tgz/.whl, a local repo checkout, a live stdio/HTTP server, or your installed MCP clients. Perfect for CI gates.
npx mcptrustchecker@latest \
@scope/mcp-server --onlineScan an npm/PyPI package, a GitHub repo or a manifest over one HTTPS request and get the full auditable report as JSON. Gate CI, power a marketplace, or embed a live badge.
curl mcptrustchecker.com/api/v1/\
scan/github/owner/repo \
-H 'Authorization: Bearer KEY'Every scan result is a link you can share, and a live Trust Score badge you can embed — no registry entry required.
Each scan has its own page and a one-click share to X, LinkedIn or Reddit (or copy the link). Post a server's A–F grade with the evidence one tap away — the receipts travel with the claim.
Grab a Markdown or HTML Trust Score badge straight from the result. It scans live and stays current — even for a server that isn't listed in the registry yet.
Tick the box while scanning to publish an npm/PyPI server to the public registry: a permanent, crawlable page and a cached, canonical badge that refreshes on every rescan.
The whole engine is on GitHub under the MIT license: the rules, the scoring model, the tests. Nothing about a grade is a black box — you can audit exactly why any server scored what it did, and reproduce it byte-for-byte.