MCP Server FAQ

What is the DomainTools MCP Server?

The DomainTools MCP Server integrates DomainTools domain intelligence directly into AI-powered workflows through the Model Context Protocol (MCP). This hosted service works with MCP-compatible AI clients to accelerate domain triage and infrastructure investigation.

Does DomainTools provide the AI model?

No. DomainTools doesn’t provide or operate the Large Language Models (LLMs). The MCP Server works with your existing AI models and agents. You control which LLM you use and how you use it.

Does DomainTools use generative AI within the product?

No. DomainTools doesn’t use generative AI to produce outputs. All tool responses are constructed programmatically from DomainTools’ proprietary databases.

Does the MCP Server comply with an open standard?

Yes. The DomainTools MCP Server implements MCP 1.2, the specification stabilized by the Linux Foundation’s Agentic AI Foundation. Using an open standard means the server isn’t tied to any single AI vendor’s proprietary protocols or security model, and any MCP-compatible client can connect to it.

How is the MCP Server delivered?

DomainTools hosts the MCP Server as a service. You connect to it over HTTPS using your MCP-compatible client — there’s nothing to install or manage on your side.

What data does the MCP Server handle?

• Input: You provide domain names and IP addresses.
• Output: DomainTools returns threat intelligence about those indicators, combining publicly available data with DomainTools’ proprietary intelligence.

Does DomainTools log or store my queries?

Yes. DomainTools logs customer queries only to investigate and resolve customer-reported support issues.

Do you use my queries to train DomainTools’ models?

No. DomainTools doesn’t train its internal predictive risk models on customer data.

How does the MCP Server protect against prompt injection?

Prompt injection — including indirect injection, where malicious instructions are embedded in external content an LLM reads — is a known risk in AI-integrated workflows. The DomainTools MCP Server addresses this in two ways:

• Input validation: The server validates all tool inputs against strict JSON schemas before execution. If a parameter doesn’t match the declared specification, the server rejects the call and returns an error. An LLM can’t be tricked into passing arbitrary commands to the server through malformed tool inputs.
• Deterministic outputs: DomainTools constructs all tool responses programmatically from its own databases — no generative AI is involved. Responses follow the tool specifications exactly and can’t be manipulated to inject instructions back into the LLM context.

Your security team controls the prompts that invoke MCP Server tools, which is the primary control point for indirect injection risk at the application layer.

Can the MCP Server hallucinate data?

No. DomainTools constructs every tool response programmatically from its own databases, so the data retrieval layer is deterministic — hallucination isn’t possible there. However, the LLM your client uses interprets the analyst’s intent and summarizes results, and LLMs can hallucinate in that layer. This is true for any LLM interacting with any MCP server — it’s a property of the model, not the data source. The DomainTools data itself is always retrieved, not generated.

What authentication methods does the MCP Server support?

The MCP Server supports two authentication modes:

• API key — pass your DomainTools API key in an X-Api-Key HTTP header. Works with every supported client. See API key authentication.
• OAuth 2.0 — pre-registered client credentials for Claude Code, Claude Desktop, Cursor, Gemini CLI, and VS Code. Your MCP client redirects you to sign in with your DomainTools account. See OAuth authentication.

During the MCP initialize handshake, the server validates your credentials against the DomainTools /v1/account endpoint.

Should I use API key or OAuth?

Use API key authentication when you want the fastest path to a working connection, or when your client isn’t in the OAuth pre-registration list (for example, Codex or Goose Desktop). Use OAuth when you want per-user sign-in tied to DomainTools accounts, or when your security policy prefers token-based flows over static headers. Both modes have equal access to MCP Server tools.

Does the MCP Server support dynamic client registration?

No. OAuth clients must be pre-registered with DomainTools. Contact your DomainTools representative or email enterprisesupport@domaintools.com to request credentials. See OAuth authentication for the registration workflow.

How does the MCP Server handle encryption?

• Data in transit: All client connections to the MCP Server use TLS 1.3-encrypted HTTPS. The server doesn’t support STDIO (Standard Input/Output) transport — STDIO is a local transport that requires the server process to run on the same machine as the client, which isn’t suitable for a remotely hosted service.
• Data at rest: You’re responsible for encrypting data at rest on your own systems, if required.