AI Context for internet infrastructure, over one MCP endpoint
Connect any MCP client once to a single endpoint, then ask about internet infrastructure in plain language. Your agent picks a read-only tool, runs it against WhisperGraph, and gets structured JSON back. Connect in the box below; the tools you get are listed in the features underneath.
Point your client at the endpoint, then ask in plain language. The agent picks the right read-only tool and answers.
claude mcp add --transport http whisper-graph https://mcp.whisper.security \ --header "Authorization: Bearer YOUR_API_KEY"
One connection, the whole graph
MCP for AI agents
A Model Context Protocol server your agent connects to once, then asks infrastructure questions in plain language — with a provably read-only query surface.
Read the overviewSeven tools, all read-only
Run Cypher, discover the schema on demand, score indicators, pull docs, attribute a batch of hosts, and run whole investigations from a shared gallery. Every one is read-only, so nothing your agent does can change anything — the worst outcome of a bad tool call is a wrong answer you can trace to its query. Every deployment advertises the same seven, so tools/list is the contract wherever you connect.
See the tool referenceConnect in one line
Point any MCP client at mcp.whisper.security with an OAuth 2.1 sign-in or a Bearer API key, and start querying. Sessions last ~6 months and survive deploys.
Set up the connectionRead-only, provably
Any write or admin Cypher is rejected before it reaches the database — even under an EXPLAIN prefix — and so is any mutating procedure, by name. There is no write tool on this server, no write scope, and no deployment where one appears, so you can hand an agent the whole surface without first working out what it might change.
How safety worksA workflow & recipe gallery
A shared library of investigation playbooks, the same one behind the use cases and the console. Agents find them with list_workflows and run them with run_workflow.
Browse the galleryThe whole internet, queryable
WhisperGraph maps over 7.4B nodes and 39.4B edges of DNS, BGP, WHOIS, certificates and threat intel from 76 feeds across 31 categories. That is 40 node labels and 50 edge types.
Threat-intel use cases
Three things that make it agent-native
Token-efficient by design
“40 labels / 50 edges — discovered on demand, never dumped into your context.”
The schema and docs load only when the agent needs them, through explain_schema and read_docs. The always-listed resource surface stays small, and a single run_workflow call does the work of several query round-trips.
One-call investigations
“Investigations in one call, not five.”
run_workflow runs a whole multi-step pivot in one call. Resolve DNS to an IP, the IP to its ASN, then check threat intel, all from one named playbook instead of a chain of hand-written queries.
Evidence for every claim
“Every answer ships the query and the rows behind it.”
Each query and run_workflow result comes back with the exact Cypher it ran, the row counts, and the timing. So when the agent makes a claim, you can see the query that backs it.
The seven MCP tools
The agent chooses among these seven on its own. Every one is read-only and returns structured rows — there is no eighth tool and no write scope, so the surface you grant is the whole surface. See the full tool reference.
queryRun a Cypher query against the graph and get back columns, rows, statistics, and an evidence block. Supports table/graph/compact formats and count-first pagination. This is the primary tool.
explain_indicatorGet a threat verdict for an IP, hostname, CIDR or ASN, with a score, a severity level, and the reasoning and sources behind it.
explain_schemaThe schema, on demand. No argument returns the label catalogue; pass a label to get its properties, edges, and a runnable sample traversal — never dumped into context up front.
read_docsPull the Whisper docs on demand — list, search, or fetch one page as Markdown — so the Cypher reference and cookbook stay out of the always-on context.
list_workflowsSearch the shared workflow & recipe gallery. Each item comes back with its summary and full parameter space, so the agent can run any variant.
run_workflowRun one or more gallery playbooks by slug in a single call — a multi-step investigation collapsed into one tool call, returning chained per-step results, derived signals, and an evidence trail.
identifyWhose infrastructure a set of hostnames belongs to — vendor and role attribution, deliberately not a threat verdict. A host it cannot place comes back with its structural neighbourhood instead of an error.
Full input schemas and example calls in the MCP tool reference.
Give your agent eyes on the internet
A free key connects any MCP client to the live graph in one step. Ask in plain language and get sourced, structured answers back.