Frequently Asked
Questions

Whisper maps the internet as one connected graph and uses it to decide what your machines and AI agents may reach. The graph joins routing, DNS, hosting, ownership, certificates and intelligence feeds into linked records you can query from any domain, IP or network. Two products run on it. Whisper Graph XDR gives each machine an identity, applies policy to where it connects, and cuts it off from outside the host. Whisper Intelligence tells you who is on the other end of any address.

Whisper runs its own graph engine. It takes in routing, DNS, registration, certificate and feed data and links it into one graph of hosts, addresses, networks and owners. Whisper Intelligence lets you ask that graph questions through the API, the Console, the MCP server or your SIEM. Whisper Graph XDR uses the same graph to decide, at each lookup, whether an enrolled machine may reach a destination. The graph runs in the EU.

Security teams that need to know what is on the other end of a connection. That means SOC analysts enriching alerts, threat hunters mapping infrastructure, and brand teams tracking look-alike domains. It also means teams that need to control where machines and AI agents connect. Agents reach it through the MCP server.

A feed tells you whether an indicator was reported. Whisper shows what it is connected to: the network that routes it, the company that owns it, the certificate it presents and the other domains its owner runs. That takes you from one alert to the rest of a campaign without switching tools. Keep your feeds. Whisper adds the context around them.

No, they are completely unrelated. Whisper Security is an internet-infrastructure intelligence company. OpenAI Whisper is an open-source speech-to-text model. The names are coincidental.

Hostnames, IPv4 and IPv6 addresses, prefixes, networks (ASNs), registration and ownership records, certificate-transparency records, DNSSEC state, mail records, physical infrastructure and feed listings. Also the links between them, such as RESOLVES_TO, ANNOUNCED_BY, REGISTERED_BY, NAMESERVER_FOR and LISTED_IN. Every entity type is listed in the graph schema reference.

An anchored query, one that starts from a name, an address or a network you already know, returns in single-digit milliseconds, measured on the server. Time over the network comes on top. The engine keeps garbage collection off the query path, so response times stay steady.

The query API takes one Cypher query as JSON and returns JSON: the columns, one row per result with the values you asked for, and a statistics block. Any HTTP client works, so you need no SDK. You can also query from the Whisper CLI or the Console.

Yes, for registration and routing. whisper.history() returns dated WHOIS snapshots for a domain and BGP routing history for an address, prefix or network. That shows who held a domain, or which network announced a prefix, at an earlier date. The rest of the graph describes the internet as it is now.

Yes. Whisper runs in our EU cloud by default. It also runs as a dedicated cloud, on premises, or air-gapped with signed graph deliveries. Talk to us about which fits.

Yes. Every account starts free, with the query API, the Console and the MCP server. Whisper Graph XDR also has a free tier that does not expire. The pricing page lists what each plan includes.

Not for the query API. Its keyless tier answers a Cypher query sent with no key, with real rows from the live graph. A key identifies your traffic, and the MCP server, the examples you run on this site and the calls whose pages say so all need one. Generate it in the Console after signing up; the MCP server takes the same key as a bearer token.

Yes. Send a Cypher query to graph.whisper.security/api/query with no key and you get real rows back. Running an example on this site takes a free account, and the MCP server always takes a key. A free account covers the API, the Console and the MCP server, and takes under a minute.

A graph database stores entities as nodes and the relationships between them as first-class edges. Traversals — "everything connected to X within N hops along these edges" — are cheap, where the same query in a relational database would force expensive recursive joins. Internet infrastructure is fundamentally a graph problem, which is why Whisper is built on one.

Cypher is purpose-built for graph queries. Its pattern syntax ((a)-[:EDGE]->(b)) maps directly to how analysts think about pivots. SQL forces graph problems through self-joins that get expensive fast. Most language models already write Cypher, and the MCP server corrects common slips, such as a wrong label name, before a query runs.

Yes. General-purpose graph databases could not answer at internet scale in single-digit milliseconds. The engine keeps its data off the garbage-collected heap. It has native types for IPv4, IPv6, prefixes and networks, so it never treats an address as a string. The engine page explains how it works.

BGP routing data, DNS observation, WHOIS and RDAP registration records, DNSSEC state, certificate transparency logs, a link graph from Common Crawl, GeoIP data, physical infrastructure records, and 134 intelligence feeds (112 of them threat feeds) across 33 categories. Most sources update continuously. A few, such as WHOIS and physical infrastructure, refresh on a slower schedule.

The default cloud deployment runs in the EU (Germany), under European data jurisdiction. Dedicated cloud and on-premises deployments are available in other regions on request.

Any MCP client — Claude Desktop, Cursor, VS Code, Continue, Windsurf, custom MCP-aware agents — can connect to the Whisper MCP server at mcp.whisper.security. The agent gains real-time access to the knowledge graph and can run Cypher queries, pivot between entities, and produce investigation reports.

On Claude, add WhisperGraph from the Connectors Directory and sign in. In Cursor, VS Code and other MCP clients, add https://mcp.whisper.security with your API key, or sign in where the client supports it. The MCP setup docs have the config for each client.

Splunk (Enterprise and Cloud), Microsoft Sentinel and Wazuh have native apps. Whisper also connects to OpenCTI and MISP on the threat-intel side, and to n8n for automation. Anything else can call the query API. The integrations page lists each one and where to get it.

Yes. It is in the Claude Connectors Directory as WhisperGraph, so on Claude you connect with a sign-in and no URL. Any other MCP client can connect to mcp.whisper.security with an API key or an OAuth sign-in.

Yes. The Whisper CLI runs a local MCP server over stdio with whisper mcp. It serves the same graph tools as the hosted server, plus one tool per recipe. For most people the hosted server is simpler.

Follow the address up its ownership chain instead of trusting a single WHOIS lookup. An IP sits inside an announced prefix, that prefix is announced by an ASN, and the ASN is registered to an organisation at a regional registry. Each step has its own record and they disagree often, because the party that registered an address block is frequently not the party using it. Whisper holds all four as linked records, so one query returns the chain and where it contradicts itself.

Pivot on what the domains share, not on the company name. Domains registered by one organisation usually share a registrant email or phone in WHOIS, a nameserver set, a TLS certificate, or a hosting prefix, and each of those is a link you can follow back out to the rest. Name search misses anything registered under a subsidiary, a product brand, or a privacy proxy. Whisper runs this from one seed domain as a prepared infrastructure-mapping workflow.

Look for infrastructure they share that is expensive to change. A common registrant email or phone, the same nameserver set, a reused TLS certificate or its fingerprint, and co-location in one small prefix are each weak alone and strong together. A shared address range at a large hoster means little; a shared certificate and registrant phone means a great deal. Whisper holds every one of those as a link, so the overlap is a traversal rather than a spreadsheet.

Generate the look-alikes, then check which ones exist. Permutation algorithms produce the plausible variants of a brand: character swaps, omissions, homoglyphs, alternative TLDs, hyphenation. Each candidate is then tested for registration, DNS resolution, mail records and certificate issuance. A registered variant with an MX record and a certificate is being prepared for use. An unregistered one is not a threat yet. Whisper runs 14 permutation algorithms and checks every result against the live graph.

BGP hijacking is one network announcing a route to address space it does not own, so traffic meant for the real owner is delivered somewhere else. You detect it by comparing what is announced against what should be: the RPKI record authorising an origin, the prefix’s announcement history, and whether a more specific route has appeared from an unexpected network. Whisper keeps RPKI state and announcement history in the same graph, so that comparison is a single query rather than three tools.

No. An absence of listings means nobody has reported it, not that it is benign. Feeds lag registration by days or weeks, and infrastructure built for a single campaign is often retired before it is ever listed. A verdict only means something with its coverage stated: which feeds were consulted, when they last updated, and whether they cover that indicator type at all. Whisper returns verdicts qualified by that coverage instead of a bare clean or dirty.

Because the name is not the destination. A filter that blocks names misses a host that connects to an address directly, uses another resolver or DNS-over-HTTPS, or looks up a name that was clean when the list was built. Whisper checks the address a name resolves to, at the moment of the lookup. A machine enrolled in Graph XDR also sends its traffic from its own address on Whisper’s network, where an egress rule can stop a connection the resolver never saw.

Constrain what the agent’s host is allowed to resolve and reach, rather than trying to constrain its prompt. An agent runs as a process on a machine with an identity, and every outbound connection it makes has a destination with an owner, a country, a route and a reputation. A policy that refuses resolution to destinations outside those bounds holds whatever the agent was talked into attempting. Whisper applies this per workload identity, with a byte ceiling per agent.

Follow the listing through to the addresses, and re-check at the moment of the connection. A sanctions list names organisations and sometimes domains, but traffic goes to addresses, and the address a name answers with changes. Blocking the name leaves the address reachable, and blocking yesterday’s address leaves today’s open. The rule has to be evaluated against what the name resolves to now and where that address lands. Whisper does this per answer.

Act somewhere the machine does not control. If containment depends on an agent running on the host, an attacker with enough privilege can stop it and the host stays on the network. Revoking the machine’s identity and tearing down its egress from the control plane removes its reach whether or not anything on the host is still answering. Whisper does both off host, with a blast-radius preview before the action and a reinstate control after it.