MCP Spotlight: WhisperGraph — 7.39B-Node DNS/BGP/Threat-Intel Graph MCP With 6 Tools & Open Cypher Query

Server: whisper-graph-mcp by Whisper Security Graph size: 7.39B nodes · 39B edges · License: Self-hostable, free open data tier Tools: 6 (read-only) · Resources: 6 · Prompts: 8 Coverage: DNS · BGP · WHOIS · GeoIP · DNSSEC · SPF · certificates · 40 threat feeds MCP Tracker: glama.ai/mcp/servers/whisper-sec/whisper-graph-mcp Docs / GitHub: github.com/whisper-sec/whisper-graph-mcp

The hardest part of any threat investigation isn't finding one IP. It's stitching together DNS, BGP, WHOIS, GeoIP, certificates, ASNs, routing paths, and 40 threat feeds into a coherent picture of who owns what, what talks to what, and what looks suspicious. Most security teams glue this together with a dozen API calls and Python joins. The result: hours of investigation per incident, missed cross-feed correlations, and an agent that can read one feed at a time but not reason across them.

WhisperGraph MCP gives your agent a single knowledge graph of the entire internet's infrastructure — 7.39 billion nodes, 39 billion edges, one schema, one query language (Open Cypher), six read-only MCP tools, six resources, eight pre-built prompts. The agent reasons across DNS, BGP, WHOIS, certificates, threat reputation, and routing paths in a single Cypher query.

The Six Tools

Tool	What It Does
run_query	Execute a Cypher query against the graph (read-only)
get_schema	Return the current graph schema — node types, edge types, properties
assess_threat	Risk assessment for an IP, domain, or ASN — pulls reputation across all 40 feeds
whois_lookup	WHOIS record for a domain or IP with structured fields
dns_lookup	DNS resolution with DNSSEC chain, record types, TTLs, nameserver topology
trace_route	BGP routing path between any two IPs/ASNs

The killer combination: assess_threat doesn't just return a score. It runs a Cypher query that pulls the IP's neighbors in the graph — what other IPs does it talk to, what domains point to it, what ASNs route through it, what's the reputation across the 40 feeds — and returns a unified picture.

The Six Resources

Resource	What It Is
schema://graph	Live graph schema for the agent to introspect
feeds://threat-intel	List of all 40 threat feeds with refresh cadence and coverage
stats://graph	Current node/edge counts, growth rate, freshness
examples://cypher	Pre-built Cypher query templates for common investigations
coverage://geographic	Geographic coverage map — which countries and regions are well-mapped
changelog://graph	Recent schema changes, new feed additions, breaking changes

Resources are the MCP pattern that lets the agent learn the schema before querying. Without schema://graph, the agent would have to guess the node types and edge labels. With it, the agent introspects the live schema and constructs accurate queries.

The Eight Prompts

Pre-built investigation workflows the agent can invoke directly:

Prompt	What It Does
investigate-ip	Full IP investigation — neighbors, threats, ownership, geolocation
investigate-domain	Domain forensics — DNS, WHOIS, certificates, subdomains, threat reputation
trace-c2-channel	Hunt for command-and-control infrastructure patterns
find-related-domains	Pivot from a known indicator to find unregistered / parked / related domains
map-infrastructure	Visualize an org's full internet footprint
detect-shadow-it	Find unsanctioned cloud assets belonging to your organization
validate-blocklist	Verify a list of indicators against current threat feeds
discover-lookalike-domains	Find typosquat / homoglyph domains targeting your brand

The prompts are the "templates" layer — well-trodden investigation paths the agent can call without writing a Cypher query from scratch. The agent can still write custom queries, but the prompts cover 80% of routine work.

The Coverage: What's in 7.39B Nodes

Data Layer	What It Maps
DNS	All registered domains, subdomains, nameservers, glue records, zone cuts
BGP	ASN relationships, routing paths, prefix announcements, peerings
WHOIS	Registrant data, registrar, creation/expiry dates, contact records
GeoIP	Country, region, city, ISP, organization, datacenter mapping
DNSSEC	Chain of trust, DS records, DNSKEY, signing status
SPF	Sender Policy Framework records, authorized senders
Certificates	CT logs, certificate transparency, SANs, issuers, validity
Threat feeds	40 sources — abuse.ch, PhishTank, URLhaus, Spamhaus DBL, AlienVault OTX, GreyNoise, etc.

The 40-feed threat reputation layer is what makes assess_threat actionable. A single IP that appears on 3 of 40 feeds is suspect. An IP that appears on 18 of 40 feeds with high confidence, hosted on a known bulletproof ASN, with rapidly rotating DNS records, is a C2 candidate.

Why Open Cypher Matters

The query language is Open Cypher — the same Neo4j-compatible graph query language that ArcadeDB, Memgraph, and most modern graph databases speak. The agent writes standard Cypher:

MATCH (ip:IPv4 {address: '203.0.113.42'})-[:RESOLVES_TO]->(d:Domain)
      -[:HAS_CERT]->(cert:Certificate)
      -[:ISSUED_BY]->(ca:CertificateAuthority)
WHERE cert.not_after < date() + duration({days: 30})
RETURN ip.address, d.name, cert.not_after, ca.name
ORDER BY cert.not_after ASC
LIMIT 50

This is the same Cypher syntax that any graph-skilled LLM already knows. The agent doesn't need to learn a WhisperGraph-specific query language. It writes the query it would write for any property graph.

Self-Hostable, Free Tier Available

The MCP server is self-hostable — your team runs the binary against your own copy of the graph. There's also a free tier that exposes the public-data portion of the graph (DNS, BGP, WHOIS, GeoIP, certificates) without the commercial threat feeds.

For threat intel teams with strict data-residency requirements (DACH/EU regulated, defense, finance), the self-hostable design means:

• The graph runs on your hardware, in your VPC
• The MCP server connects to your local graph
• Your agent queries never leave your perimeter
• Threat intel from commercial feeds is added via your existing feed contracts

The architecture is the same as running Neo4j + a thin MCP bridge, except the graph comes pre-loaded with 7.39B nodes you don't have to ingest yourself.

Facio Integration

{
  "mcpServers": {
    "whispergraph": {
      "url": "https://your-whispergraph-instance.example/mcp",
      "headers": {
        "Authorization": "Bearer ${credentials.WHISPERGRAPH_TOKEN}"
      }
    }
  }
}

Facio's audit trail captures every threat investigation the agent runs: which IPs and domains were queried, what Cypher statements were executed, what threat assessments were returned, and what actions followed. For security operations, this creates a complete investigative record — perfect for incident retrospectives, regulatory evidence (DORA, NIS2, BAIT), and team handoff.

For HITL workflows, the entire MCP surface is read-only by design — run_query rejects any non-MATCH statement, assess_threat doesn't take actions, trace_route doesn't send probes. The agent investigates; the human takes the response action (block IP, sinkhole domain, open ticket). The MCP is the read-side; the human is the response-side; Facio captures both.

Quickstart

# 1. Deploy WhisperGraph (self-hosted)
docker run -d \
  -p 7687:7687 -p 7474:7474 \
  -v whispergraph-data:/data \
  whisper-sec/whispergraph:latest

# 2. Install the MCP server
npm install -g @whisper-sec/whisper-graph-mcp

# 3. Add to your MCP client
{
  "mcpServers": {
    "whispergraph": {
      "command": "whisper-graph-mcp",
      "env": {
        "WHISPERGRAPH_URL": "bolt://localhost:7687",
        "WHISPERGRAPH_USER": "${credentials.WHISPERGRAPH_USER}",
        "WHISPERGRAPH_PASSWORD": "${credentials.WHISPERGRAPH_PASSWORD}"
      }
    }
  }
}

# 4. First prompts
# "Investigate 203.0.113.42 — what does it talk to, who's responsible, what's the threat posture?"
# "Trace the BGP path from our office IP to a known-malicious ASN"
# "Find all lookalike domains targeting our brand — homoglyphs and typosquats"
# "Detect shadow IT — find all AWS, GCP, Azure assets belonging to our organization"
# "What domains point to IPs that also host known C2 infrastructure?"

Use Cases

Incident response: "We're seeing beaconing from a finance subnet to 198.51.100.66 every 30 minutes. Investigate." Agent runs assess_threat + whois_lookup + a Cypher query for neighbors + DNS history. Returns a structured threat brief with indicators, owner, and reputation.

Threat hunting: "Find all domains registered in the last 30 days that have an SPF record pointing to a sender our org doesn't recognize." Cypher query with time-window and pattern match.

Brand protection: "Find all lookalike domains for ourbrand.com — character substitution, TLD swaps, hyphenation, homoglyphs." Runs the discover-lookalike-domains prompt.

M&A due diligence: "Map the full internet footprint of this acquisition target — what domains, IPs, ASNs, certificates do they own?" Runs map-infrastructure and produces a comprehensive asset inventory.

Shadow IT discovery: "We own the ASN and IP range. Find all cloud-hosted assets in our space we don't know about — AWS, Azure, GCP, DigitalOcean, Hetzner." Cypher query against the BGP and hosting-provider graph.

Supply chain risk: "For each of our 200 SaaS vendors, find what subdomains they have, what certificates they've issued, what hosting providers they use. Flag any vendor with infrastructure recently moved to high-risk ASNs." Mass investigation with Cypher joins.

Bottom Line

WhisperGraph MCP is the first internet-scale infrastructure graph your agent can query directly through MCP. 7.39 billion nodes, 39 billion edges, 40 threat feeds, Open Cypher, six read-only tools, six resources, eight investigation prompts. Self-hostable for regulated environments, free tier for the open data.

For security teams, this is the missing layer that turns "agent reads one feed at a time" into "agent reasons across the entire internet infrastructure graph in one query." The threat investigation workflow that used to take an analyst 4 hours of Python joins now runs as a single MCP call.

docker run whisper-sec/whispergraph:latest and your agent sees the entire internet.

---

MCP Spotlight is a series covering servers that give AI agents real capabilities. Every server is evaluated for graph coverage, query expressiveness, and integration fit with Facio's HITL-first agent runtime.