threatcheck

threatcheck

threatcheck puts a threat-intelligence gate in front of the things an agent is about to touch. Before a fetched page, email, ticket or tool result gets acted on, it can be scanned for prompt injection and for the URLs, domains and IPs buried in it. Before a link gets opened, a single indicator — an address, a hash, an email, a phone number — can be looked up for a reputation verdict with a risk score and the blocklists citing it. CVEs can be read in detail, listed fresh off the press, or listed by confirmed exploitation in the wild. And a site you run, or a project's dependency file, can be put under daily watch so new findings arrive as they open.

This package decides nothing locally. It answers the tool list itself, exactly as captured from the live endpoint, and relays each tool call upstream, handing back whatever returns untouched. Several calls are metered against your own monthly quota, per the tool descriptions below, and the watch tools spend watched sources on your own plan. There is no tool here that buys anything, tops a plan up, checks out, or mints a key: the surface only spends against an account you already hold, and the finding tool only changes the status of findings inside your own organization.

Tools

• scan_before_use — scan untrusted content for prompt injection and check the URLs, domains and IPs it contains before acting on it. Required: content. Optional: source_url. Returns a verdict of block, warn or allow.
• check_url — the cheap pre-fetch gate for one URL, domain or IP. Required: url. Returns block, warn or allow.
• check_indicator — the full reputation picture for one indicator: an IP, domain, URL, file hash (MD5, SHA-1 or SHA-256), email address or phone number, with defanged forms accepted. Required: indicator. Optional: enrichment (fast, standard, basic) and country for national-format phone numbers. Returns a verdict, a headline, a recommended action, a 0–100 risk score, citing blocklists, evidence, first/last seen and context. Costs one request of the monthly quota.
• get_cve — one CVE in full: description, severity, CVSS, EPSS, CISA KEV status, exploitation evidence, affected packages and vendor advisories. Required: id (for example CVE-2021-44228). Costs one request of the monthly quota.
• recent_cves — the most recently published CVEs, newest first, at most 20. Optional: severity (CRITICAL, HIGH, MEDIUM, LOW) and limit (1–20, default 10). Costs one request of the monthly quota.
• search_indicators — find listed domains that look like a brand or domain: typosquats, homoglyphs, the same name on other TLDs, phishing-word wrappers. Required: keywords. Optional: limit (1–500, default 50). Costs one request of the monthly quota.
• check_indicators — reputation for up to 100 indicators in one call, mixed kinds, defanged forms accepted. Required: indicators. Optional: enrichment (basic, standard, full) and country. Every unique indicator sent costs one request of the monthly quota.
• scan_email — scan one email message for phishing and malware before delivering or opening it. Send exactly one of eml (the raw RFC 5322 message) or message (the fields you already parsed). Optional: authservId, trustAuthenticationResults, trustedHops, connectingIp. Returns a verdict, a recommended action, a risk score, reasons, and what could not be checked. Costs one scan of the scan meter per message.
• exploited_cves — CVEs with attested exploitation in a recent window. Optional: days (1–30, default 7) and limit (1–50, default 20). Costs one request of the monthly quota with a key; works without one, rate-limited by IP.
• watch_domain — start daily vulnerability monitoring of one public website you run. Required: domain. Optional: name. Uses one watched source of the plan; re-adding a watched domain costs nothing.
• watch_dependencies — start vulnerability monitoring of a project from the text of its lockfile or SBOM (pass the content, not a path). Required: content and name. Optional: format and includeDev. Uses one watched source of the plan.
• ack_finding — set the status of one of your own monitoring findings: ACKNOWLEDGED, PATCHED, NOT_AFFECTED, or SNOOZED (with snoozedUntil, at most 90 days ahead). Required: findingId and status.

An email address or phone number checked through check_indicator or check_indicators gets a threat verdict about that indicator — malicious, suspicious, clean or unknown, with sources — never personal data about a person, and it is never called clean merely because nothing is known about it.

Configuration

• ISMALICIOUS_MCP_URL (optional) — override the upstream endpoint. Defaults to https://api.ismalicious.com/mcp. (The override hook is this rebuild's own labelled reconstruction.)
• ISMALICIOUS_API_KEY (optional) — when set, every upstream request carries an Authorization: Bearer <key> header holding your own isMalicious API key. When it is not set, no key header is sent and calls run keyless. (The env name is pinned by the original record's npm package entry and the header name by its remote declaration; the Bearer form and the forwarding itself are labelled reconstructions — the original surface was captured keyless.)

Forwarding mechanics

The first tool call opens a streamable HTTP session upstream: a JSON-RPC initialize request (protocolVersion 2025-03-26, clientInfo name threatcheck), then a notifications/initialized notification, then the tools/call request with the caller's arguments unchanged. When the upstream answers initialize with an Mcp-Session-Id header, that value rides on every later request, and the session is kept for the life of the process. Replies may arrive as plain JSON or as server-sent events; both are read.

Successful upstream results pass through verbatim, including any isError flag the upstream sets. Failures — a JSON-RPC error, a non-2xx status, an unreadable response, or a request that cannot be completed — come back as an isError result whose text starts with threatcheck error: (this rendering is this rebuild's own labelled reconstruction).

CLI

Each tool is also a subcommand taking one JSON argument string:

npx @reneturcios/threatcheck check_url '{"url":"https://example.com"}'
npx @reneturcios/threatcheck check_indicator '{"indicator":"203.0.113.7"}'
npx @reneturcios/threatcheck get_cve '{"id":"CVE-2021-44228"}'
npx @reneturcios/threatcheck --version

Exit codes: 0 on success, 1 on an isError result, 2 on a usage error.

MCP server

npx -y @reneturcios/threatcheck mcp

MCP client config:

{
  "mcpServers": {
    "threatcheck": {
      "command": "npx",
      "args": ["-y", "@reneturcios/threatcheck", "mcp"]
    }
  }
}

Docker

docker run -i ghcr.io/cloudcorprecords/threatcheck
docker run --rm ghcr.io/cloudcorprecords/threatcheck --version

The container's default command is mcp, so the first form is a stdio MCP server; any other arguments reach the CLI normally.

Development

npm install
npm run build
npm test
npm run dev

The test suite runs entirely against a stubbed JSON-RPC upstream — no live network calls.

License

MIT — see LICENSE.