See the tools work,
safely.
Hybrid demo Exactly two MCP tools exist. KeySweep runs here live in a rate-limited sandbox — the same scanner the qualified keysweep_scan_text tool uses, with the same redacted output. MicroDefend is a guided example with sample output, because a public page must never become an arbitrary network scanner. No account, no credential, and no plan allowance is involved anywhere on this page.
Try it.
KeySweep analyzes text you supply for exposed secrets — API keys, tokens, private keys, credentials, database URLs. The analysis is pure pattern detection: no network access, no AI call, nothing leaves the process. The response is redacted by design: secret values are never returned, only their type, severity, confidence, a fingerprint, and a bounded value length.
Use sample or non-sensitive text. Do not paste live production credentials.
Nothing yet — run the sandbox to see a live redacted result.
Sandbox result Live analysis of your input by the production scanner, in this page's isolated demo path. Not a plan-allowance call.
- status
- risk_level
- confidence_score
- text_length
| type | severity | confidence | fingerprint | value_length |
|---|
- Validated first. Empty or oversized input is rejected before anything runs — and, like the production tool, a rejected request costs nothing.
- Rate limited. The demo is limited per IP address (5 runs per hour, 20 per day). If the rate-limit store cannot be verified, the demo fails closed rather than running ungoverned.
- Scanned in-process. The same pure scanner behind
keysweep_scan_textanalyzed your text. No network access, no external API, no credential. - Redacted. The service's own redaction mapped raw detections to type, severity, confidence, fingerprint, and bounded length. Your text was never stored or logged — only its length and the detection count.
MicroDefend scans a public website you are authorized to test: TLS posture, security headers, metadata, and an exposed-path audit. Every fetch is SSRF-defended — targets are validated and pinned to resolved addresses before any connection, redirects are re-validated per hop, and response bodies are size-capped.
This demo does not contact the URL. A public page that scanned arbitrary URLs could be abused as a network scanner — so the walkthrough below uses a labeled sample result instead.
Sample result Illustrative output in the qualified response shape. No scan ran; no target was contacted.
- Target validation first. The hostname resolves via DNS, and every returned address is checked before any connection: private, loopback, link-local, reserved, multicast, and cloud-metadata destinations are rejected outright.
- Constrained transport. HTTP/HTTPS only, standard web ports (80/443) only, no credential-bearing URLs, URLs capped in length.
- Pinned connections. The scan connects to the validated address, not a re-resolved one. Redirects are never followed blindly — each hop is re-validated and re-pinned, with a redirect cap.
- Bounded work. Response bodies are size-capped (2 MiB) and HTTP fetches carry connect/read timeouts. Note the qualified metering asymmetry: a MicroDefend call consumes allowance before target validation, so a rejected target may still consume allowance.
What this demo is — and is not.
- Two tools, no more. The public MCP registry exposes exactly
keysweep_scan_textandmicrodefend_scan_website. This page demonstrates nothing else. - Sandbox, not allowance. KeySweep runs here do not consume plan allowance and require no credential. Production calls authenticate with an issued MCP bearer credential and consume metered allowance under fixed UTC minute/hour/day windows.
- Your input stays out of the record. Submitted text is never stored, never logged, and never sent to analytics. The only record is an anonymous per-IP counter enforcing the demo limits. Demo submissions are otherwise not persisted.
- Not a network proxy. The MicroDefend walkthrough performs no fetch. Nothing on this page can be used to scan, probe, or reach a third-party or internal host.
Qualification boundary
Qualified: the public JWKS verification surface, and authenticated MCP end to end (post-roadmap production qualification gate, 2026-09-14). A successful sandbox run on this page is not the credentialed path — no account, credential, or allowance is involved anywhere on this page.
Not qualified, and not live: MCP billing. There is no MCP checkout, subscription, or paid plan.
Make it real.
Point an MCP-capable client at the machine endpoint with an issued credential, discover the two tools, and make the same calls for real. Pricing is not published and billing is not live; access is handled directly.