What is a domain dossier?

Best move: ask for one correlated dossier per domain, not five separate lookups you have to stitch together yourself.

Why it works: the layers explain each other. A model provider next to a vector store next to a firmographic vertical is a lead. The same facts in five disconnected tabs are trivia.

Key takeaways

• A domain dossier is a single report that correlates five layers of evidence about one domain, not a single-layer scan.
• The five layers are tech stack, infrastructure, firmographics, AI-readiness, and security.
• The top-level summary fuses those layers into a primary stack, an AI-readiness verdict, a security grade, and a hot, warm, or cold ICP signal.
• Correlation is the point. A model provider sitting next to a vector store and a vertical is a qualified lead that five separate tools would never connect.
• Bassethound returns one dossier per domain from a keyless, stateless MCP call, so any agent host can request it inline.

What goes in a domain dossier?

Five layers, returned together. The tech-stack layer lists frameworks, CMS, analytics, widgets, payments, and ecommerce, each detection carrying a category, version, confidence, and the evidence behind it. The infrastructure layer resolves the IP, ASN, hosting org, CDN, nameservers, MX records, email provider, and DNS TXT vendors, then reads the email-auth booleans (SPF, DKIM, DMARC) and the TLS certificate down to issuer, SANs, and days remaining. The firmographics layer names the company, its vertical, description, contacts, and social profiles. The AI-readiness layer fingerprints the backend: model providers, AI SDKs, orchestration frameworks, vector stores, LLM observability, AI widgets, llms.txt, MCP endpoints, and .well-known AI files, then scores what it finds. The security layer grades response headers A to F and pairs that with a TLS grade.

On top sits a summary that names the primary stack, states an AI-readiness verdict, prints a security grade, and flags an ICP signal. That last field is the payload. It reads all five layers at once and tells you whether the domain is hot, warm, or cold.

How is a dossier different from a WHOIS or BuiltWith lookup?

Each of those is one layer. WHOIS and RDAP return registration and infrastructure facts. BuiltWith and Wappalyzer return the tech stack. Censys, Shodan, IPinfo, and DomainTools return infra and OSINT. Apollo, ZoomInfo, and Clearbit return firmographics. Every one of them is good at its slice, and each has a real corpus behind it. A dossier is different in shape, not just in breadth. It runs those detections against the same domain in the same request, then reasons across them. The infrastructure layer already tells you the CDN and email provider, so the AI-readiness layer does not guess at hosting. The firmographic vertical gives the AI signals a meaning: a legal-tech company shipping a vector store reads differently from a marketing agency doing the same.

You could buy five subscriptions and glue their outputs together with a script. You would spend the budget, own five API contracts, and still lack the join. The dossier does the join for you, in one call, keyless.

Why correlate the layers instead of fetching them separately?

Because the answer you want lives between the layers, not inside any one of them. Fan-out gives you five files. Fusion gives you a verdict. An agent that calls five MCP servers pays five round trips, reconciles five schemas, and still writes the correlation logic itself, per domain, every time. The correlation is the expensive part, and it is the part none of the single-layer tools hand you.

Consider the ICP signal. It goes hot when the AI-readiness layer finds a shipping stack, the firmographics layer puts the company in a vertical you sell to, and the infrastructure layer shows a footprint worth crawling. No single layer produces that. It is a read across all of them at once. A competitor can mirror the tech-stack layer cheaply, because detection is commoditized. What they cannot cheaply mirror is the join, delivered inline, in the same request that produced the parts. That is the moat: one call, five correlated layers, one signal an agent can act on without doing the stitching.

How fresh is a dossier, and what can it miss?

Every dossier is built from a read of the domain rather than a prebuilt corpus, so the facts reflect that crawl. Three profiles trade depth for speed. Fast is deterministic and keyless, roughly one to three seconds, DNS and headers and cert. Standard adds the AI stack and firmographics. Deep adds follow-on probes, the security grade, and Wayback first-seen history, so you can date when a signal appeared.

Be honest about the ceiling. The engine reads a static crawl plus an optional JavaScript render. A component that runs fully server-side, sits behind a proxy, or initializes only in a browser context the render did not trigger can stay invisible. A vector store called from a backend the page never touches will not show. The dossier does not paper over this. It reports what it reached and flags the gaps, so a missing signal reads as unknown, not as absence. That distinction matters when you score intent. Treat the dossier as strong positive evidence, honest about its blind spots, never as proof of a negative.

How do you pull a domain dossier from an agent?

One tool, one call. Bassethound is a remote MCP server at mcp.bassethound.ai/mcp. Any agent host that speaks MCP (Claude, Cursor, ChatGPT) connects to the endpoint and calls sniff_domain with a domain and a profile. There is no key to provision and no crawl to run yourself. The free tier is keyless, rate-limited, because Bassethound owns the crawl and does not ask you to bring your own. Higher depth and volume run through OAuth on the paid tier.

The server is stateless and read-only. Each call is self-contained, and nothing carries from one request to the next, so there is no session to warm up. For an agent, the dossier is a single inline step in a larger investigation. Ask for the domain, get five correlated layers and a signal back, and keep going. You are not orchestrating five services or holding a corpus. You point one tool at one domain and read the answer.

Bassethound perspective

The industry sells lookups. One layer, one corpus, one subscription, and the correlation is your problem. That model made sense when a human read the output. An agent does not want five tabs. It wants a verdict it can act on inside the request it is already running. So the axis that matters is not who has the biggest tech-stack corpus. BuiltWith and Wappalyzer already won that, and mirroring a single layer is cheap. The axis that matters is the join, and the deep AI-stack read nobody else does well. Most AI-readiness scores are a homepage skim and an llms.txt check. That rewards marketing. Bassethound fingerprints the backend (model providers, SDKs, orchestration, vector stores, observability) and correlates it with the vertical and the infrastructure, so the signal reflects engineering, not copy. A competitor can clone the commoditized layer tomorrow. Cloning the fusion means running all five in one request and reasoning across them, keyless, per domain. That is the product: not a better BuiltWith, but a dossier an agent trusts as one step, and a signal that reads the whole domain at once.

Sources

• Model Context Protocol: https://modelcontextprotocol.io
• WHOIS Protocol Specification (RFC 3912): https://datatracker.ietf.org/doc/html/rfc3912
• MDN Web Security: https://developer.mozilla.org/en-US/docs/Web/Security
• llms.txt proposal: https://llmstxt.org