Binary³ MCP Tools

Security capabilities
your agents can call.

Explore the currently available Binary³ MCP tools and their verified interfaces. Every entry below mirrors the registered tool surface — nothing is listed here that a client cannot discover from the endpoint.

Connect MCP View Docs

MCP ENDPOINT — machine-facing protocol surface, not a webpage
https://mcp.binarycubed.com/mcp

Binary³ capability · KeySweep

keysweep_scan_text

keysweep_scan_text(text: string)

Inspects supplied text for exposed secrets: API keys, tokens, private keys, credentials, and database URLs. Pure analysis of the submitted content — the tool performs no network access.

  • Operation typeAnalytical. Read-only inspection of the supplied text; no network access. Scan activity is written to the service audit log on a best-effort basis; authentication and entitlement checks run before the service; failures of those checks produce security telemetry.
  • Parameterstext — required string. Must be non-empty after trimming; maximum 256,000 characters.
  • Outputstatus, risk_level, confidence_score, summary (with total, by_severity, by_type), detections (each with type, severity, confidence, fingerprint, value_length), and text_length.
  • Error behaviorExpected failures — validation, quota, binding, and scan failures inside the service layer — return a structured MCP tool error (an is_error result) whose message carries a stable, machine-readable code (the SDK prefixes tool error messages with the tool's name) — never a transport-level 500. An unexpected infrastructure failure may surface without such a code; an is_error result is a failure regardless. Argument-shape failures are handled by the SDK before the tool runs, and transport-layer authentication failures (for example a missing bearer credential) fail before any tool invocation. Codes include: invalid_text, text_too_large, rate_limit_exceeded, quota_unavailable, scan_failed, db_scope_unavailable, db_role_unavailable, binding_check_unavailable, credential_binding_failed. When a call reaches a tool without a resolved identity it yields unauthenticated; a credential without this tool's entitlement yields forbidden.
  • Security noteCredential-gated: the bearer credential must carry the keysweep_scan_text entitlement. Output is redacted — detected secret values are never returned; findings carry only type, severity, confidence, a truncated fingerprint, and the secret's length.
  • Example useScreen code, configs, logs, or messages for leaked credentials before an agent stores, forwards, or acts on them.

Documentation →   Simulated invocation →

Binary³ capability · MicroDefend

microdefend_scan_website

microdefend_scan_website(url: string)

Runs a website security scan — TLS posture, security headers, metadata, and exposed-path audit — against a public URL you are authorized to scan.

  • Operation typeAnalytical. Outbound requests are confined to the supplied URL, its redirect chain, and probe paths on the target — every request and every redirect hop is independently re-validated against the same target policy (public HTTP/HTTPS on ports 80/443 only, maximum 3 hops), and nothing on the target is modified. Completed scan results are persisted server-side; scan activity is written to the service audit log on a best-effort basis.
  • Parametersurl — required string. Maximum 2,048 characters; http or https scheme only; ports 80 and 443 only; embedded credentials are not accepted. Public targets only — private, loopback, link-local, reserved, multicast, and cloud-metadata addresses, and internal hostnames, fail validation before any HTTP request to the target is issued.
  • Outputstatus, scan_id, findings, and a redacted result.
  • Error behaviorExpected failures — validation, quota, concurrency, binding, and scan failures inside the service layer — return a structured MCP tool error (an is_error result) whose message carries a stable, machine-readable code (the SDK prefixes tool error messages with the tool's name) — never a transport-level 500. An unexpected infrastructure failure may surface without such a code; an is_error result is a failure regardless. Argument-shape failures are handled by the SDK before the tool runs, and transport-layer authentication failures (for example a missing bearer credential) fail before any tool invocation. Codes include: invalid_target, url_too_long, invalid_scheme, credentials_in_url, blocked_hostname, blocked_port, dns_resolution_failed, blocked_ip, scan_failed, scan_slots_unavailable, rate_limit_exceeded, rate_limit_unavailable, db_scope_unavailable, db_role_unavailable, binding_check_unavailable, credential_binding_failed, duplicate_scan, too_many_concurrent_scans, scanner_busy. When a call reaches a tool without a resolved identity it yields unauthenticated; a credential without this tool's entitlement yields forbidden.
  • Security noteCredential-gated: the bearer credential must carry the microdefend_scan_website entitlement. Every fetch is SSRF-defended: targets are validated and IP-pinned before any connection, redirects are re-validated per hop (maximum 3), response bodies are capped, and private or internal targets fail validation before any HTTP request to the target is issued.
  • Example useCheck the public security posture of your own properties from inside an agent workflow.

Documentation →   Simulated invocation →

Access

Credential-gated,
terms not yet published.

Both tools require an issued credential with the specific tool's entitlement. MCP plan pricing is not yet published; access conversations start with your intended tools and agent environment.

Access status

Two verified tools.
Nothing invented.

Connect MCP View Docs