IsMalicious threat intelligence for AI agents

What isMalicious threat intelligence for AI agents does

isMalicious threat intelligence for AI agents exposes threat intelligence and security checks through the Model Context Protocol. It covers reputation lookups for IP addresses, domains, URLs, file hashes, email addresses, and phone numbers; CVE research; prompt-injection detection; and phishing or malware scanning for complete email messages.

The server also includes vulnerability-monitoring actions. Agents can start watching a domain or dependency source, and acknowledge findings. These operations change account state, while the lookup tools are read-only.

How it works

A local deployment runs as the @ismalicious/mcp-server package through npx. It communicates with an MCP client over the client’s configured local process connection and uses an API key and secret for authenticated requests. Free keys are available through the isMalicious account page. If credentials are omitted, the local process can still start and exposes bootstrap_key, which creates a free key from an email address for the session.

The hosted endpoint is available at https://api.ismalicious.com/mcp over Streamable HTTP. Requests carry the credentials in an Authorization: Bearer <apiKey>:<apiSecret> header, or through the supported encoded alternatives. The endpoint is stateless: each POST contains one JSON-RPC message or a batch of up to 10 messages.

Setup and configuration

A local MCP configuration can start the package with:

{
  "mcpServers": {
    "ismalicious": {
      "command": "npx",
      "args": ["-y", "@ismalicious/mcp-server"],
      "env": {
        "ISMALICIOUS_API_KEY": "your-api-key",
        "ISMALICIOUS_API_SECRET": "your-api-secret"
      }
    }
  }
}

For remote clients, configure the MCP URL and add the authorization header. Claude Code can register the endpoint with its HTTP transport, while clients using JSON server entries can specify the URL and headers directly. The package also exports handleStreamableHttp(request, options) from @ismalicious/mcp-server/streamable-http for self-hosted web-standard Request and Response handling.

Tools and capabilities

• scan_before_use checks untrusted text for prompt injection and link reputation, returning a block, warn, or allow result.
• check_url evaluates a URL, domain, or IP before retrieval.
• check_indicator handles a single IP, domain, URL, hash, email, or phone number and returns a verdict, risk score, evidence, context, and recommended action.
• check_indicators checks multiple indicators, while search_indicators searches the isMalicious service.
• get_cve, recent_cves, and exploited_cves provide CVE details, recent vulnerability lists, and exploited-vulnerability information.
• check_password_exposure checks a password locally; scan_email analyzes one complete email message for phishing and malware.
• watch_domain, watch_dependencies, and ack_finding support vulnerability monitoring workflows.

Limitations and notes

The hosted endpoint does not expose check_password_exposure or bootstrap_key, because password scanning is intended to keep the password on the local machine and remote bootstrapping is unavailable. Remote calls require an accepted key; listing tools does not require one, but tool calls do.

Tool calls consume service usage according to their operation. Repeated results can use a shared cache for the same key. Remote clients should allow at least 60 seconds per call, and check_indicators has a shorter 50-second edge deadline. A client that stops waiting can still incur the call’s charge. The remote ismalicious://quota resource reports scan-meter information.