Skip to main content
Glama

WhisperGraph is an MCP server backed by the world's largest internet-infrastructure graph database - 46 billion nodes and edges across 20 entity types, mapping every domain, IP, ASN, prefix, organization, Web link and threat-intelligence listing into a single Cypher-queryable graph. Used by security teams, incident responders, and AI agents for investigation, attribution, brand protection, and infrastructure forensics.

Built for agents from day one.

  • Programmatic signup in 2 HTTP calls. No browser, no CAPTCHA, no human-in-the-loop. Email verification only. Working API key in ~5 seconds.

  • Free trial for everyone, including agents. Paid tiers for higher quotas.

What you can ask:

  • DNS: resolution, nameservers, MX, SPF chains, DNSSEC

  • Routing: ASN ownership, BGP origin history, MOAS conflicts, peering

  • Hosting & ownership: registrar, WHOIS contacts, organization mapping

  • Threat intel: ~40 feeds across 18 categories, CALL explain() for full threat scoring

  • Historical: WHOIS history, BGP route changes

  • Web: 10.9B hyperlinks for inter-domain analysis

Learn more: Agent signup · WhisperGraph intro · Cypher API reference · Query guide · Cypher syntax · Functions · Best practices · MCP setup

Quick start

You need a WhisperGraph API key. Get one either:

Hosted remote server (no install)

Whisper runs a hosted MCP server at https://mcp.whisper.security. Point any MCP client that supports remote servers at it and authenticate with your API key:

{
  "mcpServers": {
    "whisper": {
      "url": "https://mcp.whisper.security",
      "headers": { "Authorization": "Bearer <your_api_key>" }
    }
  }
}

MCP clients (stdio)

For local stdio transport, add this to your MCP client config:

{
  "mcpServers": {
    "whisper-graph": {
      "command": "npx",
      "args": ["-y", "@whisper-security/whisper-graph-mcp"],
      "env": { "WHISPER_API_KEY": "your-api-key" }
    }
  }
}

Self-hosting this repo is for teams who want to run the MCP layer in their own environment. See How to set up.

Related MCP server: Lookup-Domain

Agent quickstart

Get a working API key in two HTTP calls - no browser, no CAPTCHA, no waiting list.

Step 1 - start signup (Whisper emails a verification code):

POST https://console.whisper.security/api/signup
Content-Type: application/json

{"email":"your-agent@example.com","attribution":{"agent_name":"your-agent","source":"<registry-name>"}}

Step 2 - verify with the emailed code:

POST https://console.whisper.security/api/signup/verify
Content-Type: application/json

{"signup_id":"...","code":"..."}

The response contains api_key, mcp_url, dashboard_url, and docs_url. Use api_key in the MCP config snippet above. Full docs: whisper.security/docs/agent-signup.

Tools

All eight tools are read-only.

Tool

What it does

query

Execute a Cypher query against WhisperGraph. Validated against a safety rule set before it reaches the backend.

list_labels

List every node label with counts. Call it before writing a query when you're unsure which label to anchor on.

describe_label

Confirm a label exists and enumerate its property keys.

explain_indicator

Threat assessment for an IP, hostname, CIDR, or ASN - score, level, factors, sources.

whisper_history

Historical WHOIS or BGP data for an indicator.

domain_variants

Typosquatting / brand-protection variants of a domain, checked against the graph.

list_recipes

List the full whisper.security catalog of ready-made recipes (see below).

run_recipe

Run any catalog recipe by slug - a keyless direct procedure or a keyed multi-step flow.

Catalog recipes

list_recipes + run_recipe expose the entire whisper.security catalog - 29 curated recipes, no hand-written Cypher required. The vendored catalog (src/catalog/recipes.json) is generated from the canonical source with npm run sync:catalog, so it tracks the platform.

Two kinds:

  • Direct recipes (keyless). A single graph procedure that runs without a key (rate-limited): assess (threat posture), identify (vendor/operator), explain, variants, origins (CDN de-cloak), history / history-whois, walk, psl-tldplusone, psl-affiliation, asset, lookup-tor-relay, db-schema.

  • Flow recipes (keyed). Curated multi-step investigations that need an API key: attack-path, attack-surface, indicator-enrichment, infrastructure-mapping, subdomain-takeover, bgp-hijack-exposure, blast-radius, route-health, typosquat, nameserver-hijack-dns-consistency, map-supply-chain-concentration, discover-ai-agent-infrastructure, build-takedown-evidence-package, indicator, anycast-dns-root-sovereignty.

// keyless direct recipe
{ "name": "run_recipe", "arguments": { "recipe": "assess", "inputs": { "v": "185.220.101.33" } } }

// keyed multi-step flow (needs WHISPER_API_KEY / X-API-Key)
{ "name": "run_recipe", "arguments": { "recipe": "indicator-enrichment", "inputs": { "value": "github.com" } } }

Each recipe carries a docsUrl (visible in list_recipes) linking to its page under whisper.security/docs.

Resources

Six MCP resources: the full schema, the relationship map, a Cypher function reference, a query cookbook, plus live whisper://stats and whisper://quota.

Prompts

Eight investigation-workflow prompt templates: investigate-ip, map-attack-surface, compare-domains, blast-radius, threat-triage, whois-pivot, bgp-investigation, typosquat-sweep.

Self-hosting (Docker / HTTP)

For remote or team deployments, run the server over Streamable HTTP:

docker run -p 8080:8080 -e MCP_TRANSPORT=http \
  ghcr.io/whisper-sec/whisper-graph-mcp:latest

Or with Docker Compose:

docker compose up

In HTTP mode the server does not authenticate inbound requests - it relays the caller's X-API-Key or Authorization: Bearer header to the hosted WhisperGraph API, falling back to the WHISPER_API_KEY environment variable when no header is present. Put it behind your own gateway if you need access control.

Configuration

All configuration is via environment variables.

Variable

Default

Description

WHISPER_API_KEY

(none)

Your WhisperGraph API key. Get one programmatically in 2 HTTP calls or via the dashboard.

MCP_TRANSPORT

stdio

stdio for local CLI use, http for remote/Docker.

HTTP_HOST

0.0.0.0

Bind host for the HTTP transport.

HTTP_PORT

8080

Bind port for the HTTP transport.

WHISPER_ALLOWED_HOSTS

(none)

Comma-separated Host header allowlist for DNS-rebinding protection in HTTP mode. Leave empty only behind a trusted gateway.

WHISPER_DB_URL

https://graph.whisper.security

Base URL of the hosted WhisperGraph API.

WHISPER_QUERY_TIMEOUT_MS

60000

Hard per-query deadline forwarded to the API.

WHISPER_DB_TIMEOUT_MS

10000

HTTP timeout for non-query calls.

LOG_LEVEL

info

debug, info, warn, or error.

Development

npm install
npm run dev       # run from source over stdio
npm test          # unit + integration tests (no secrets needed)
npm run build     # bundle to dist/
npm run lint      # eslint
npm run typecheck # tsc --noEmit

Contributing

Contributions are welcome. See CONTRIBUTING.md and our Code of Conduct. Security issues: see SECURITY.md.

License

Apache-2.0. "Whisper", the Whisper logo, and "WhisperGraph" are trademarks of Whisper Security - see NOTICE.

Available Tools

8 tools
describe_labelDescribe WhisperGraph LabelA
Read-onlyIdempotent
Inspect

Describe a single label: confirm it exists, get its node count, and enumerate the property keys observed on that label.

Use this BEFORE writing a query that filters on a specific property. If you write WHERE h.fqdn = "..." but describe_label("HOSTNAME") returns properties = ["name", "threatScore", ...], your query will silently scan the entire label. Verify first.

Argument: label (string, required) - uppercase letters, digits, and underscores only.

Returns: {name, exists, count, properties[], edgesDoc}. Cached 5 minutes.

Tip: edge types are NOT in the response - see the whisper://schema/relationships resource for which edges connect this label to others.

Inputs: label (required). Returns: name, count, error, exists, +4 more

domain_variantsTyposquat / Brand-Protection VariantsA
Read-onlyIdempotent
Inspect

Generate typosquatting / brand-protection variants of a domain or brand name and check which ones actually exist in WhisperGraph.

Runs 14 mutation algorithms - character omission, repetition, transposition, QWERTY-adjacent replacement/insertion, vowel-swap, bitsquatting, homoglyph / Unicode confusables, hyphenation, dot insertion/omission, TLD-swap, TLD-addition, and subdomain-add. Unicode input is accepted (and expected) so IDN homoglyph lookalikes resolve correctly.

Returns { rows: [...] }. Each row: { variant, method, exists, nodeId, label, confidence (0.3-0.9), confidenceLabel (low/medium/high) }. By default only variants that EXIST as nodes are returned - the registered lookalikes worth investigating. Note that "exists" means registered/observed, NOT malicious: pivot each hit through explain_indicator for a threat verdict.

Arguments:

  • name (string, required) - the domain or brand to mutate, e.g. "google.com". Allowed characters: letters (including Unicode), digits, '.', '-', '_'.

  • label (string, optional, default HOSTNAME) - node label to check existence against.

  • includeNonExistent (boolean, optional, default false) - when true, also return generated variants that do NOT exist in the graph (larger, noisier result set).

Performance: typically <150ms. Results are capped at 500 rows.

Inputs: name (required), label, includeNonExistent. Returns: rows

explain_indicatorThreat Assessment for an IndicatorA
Read-onlyIdempotent
Inspect

Run a comprehensive threat assessment on a single indicator. The indicator can be an IPv4, IPv6, hostname, CIDR network, or ASN - the procedure auto-detects the type.

Returns a single structured row: { indicator, type, available, cached, found, score, level (NONE/INFO/LOW/MEDIUM/HIGH/CRITICAL), explanation, factors[], sources[] }. For ASN inputs the row also includes a breakdown object with composite sub-scores (threatDensityScore, graphMetricsScore, historicalScore, prefixAgeScore). For CIDR inputs the explanation field carries threat-density stats (listed IPs, density %).

Prefer this tool over manual ASN→PREFIX→IP→LISTED_IN walks - those time out on large ASNs (AWS, GCP, Azure, Cloudflare). Performance: 3-25ms for IP/domain/network, up to ~80ms for ASN.

Argument: indicator (string, required). Allowed characters: letters, digits, '.', '-', ':', '/', '_'. Cypher-special characters are rejected.

Inputs: indicator (required). Returns: rows

list_labelsList WhisperGraph LabelsA
Read-onlyIdempotent
Inspect

List all node labels in WhisperGraph with their counts.

Use this BEFORE writing a query when you're not sure which label to anchor on. It rules out hallucinated labels (e.g. there is no DOMAIN or FQDN - only HOSTNAME) and tells you which labels are large (HOSTNAME, IPV4) vs small (RIR, COUNTRY).

Returns: an array of {label, count} rows. Cached server-side for 5 minutes.

Tip: pair with describe_label to verify which properties exist on a label before referencing them in WHERE clauses.

Inputs: no named parameters. Returns: labels

list_recipesList WhisperGraph Catalog RecipesA
Read-onlyIdempotent
Inspect

List the whisper.security catalog of ready-made recipes - the full set exposed by this server. Two kinds:

  • direct recipes (keyless) are a single graph procedure: whisper.assess (threat posture), whisper.identify (vendor/operator), whisper.explain, whisper.variants (typosquats), whisper.origins (CDN de-cloak), whisper.history / whisper.history.whois (WHOIS timeline), whisper.walk, whisper.psl.*, whisper.asSet, whisper.lookupTorRelay, db.schema. These run WITHOUT an API key (rate-limited).

  • flow recipes (keyed) are curated multi-step investigations: attack-path, attack-surface, indicator-enrichment, infrastructure-mapping, subdomain-takeover, bgp-hijack-exposure, blast-radius, route-health, typosquat, nameserver-hijack-dns-consistency, map-supply-chain-concentration, discover-ai-agent-infrastructure, build-takedown-evidence-package, indicator, anycast-dns-root-sovereignty. These need a WhisperGraph API key.

Each entry returns { slug, title, purpose, category, mode, access, requiresKey, inputs[], params[], columns[], docsUrl }. Run any of them with run_recipe(recipe=, ...). Optional filters:

  • mode ("direct" | "flow")

  • access ("keyless" | "keyed")

The catalog is generated from the canonical whisper.security catalog, so this list stays in sync with what the platform ships.

Inputs: mode, access. Returns: recipes

queryWhisperGraph Cypher QueryA
Read-onlyIdempotent
Inspect

Execute a Cypher query against WhisperGraph - the internet's largest infrastructure graph database (7.39B nodes, 39B edges, 5.6M threat intel edges). Returns JSON with columns, rows, and statistics.

Use this tool for any question involving domains, hostnames, IPs, DNS, BGP, GeoIP, web links, email infrastructure, WHOIS, DNSSEC, or threat intelligence.

NODE LABELS (20): HOSTNAME (2.6B), IPV4 (619M), IPV6 (820K), PREFIX (2.5M), ASN (116K), ASN_NAME (108K), ORGANIZATION (119M), CITY (54K), TLD (1.7K), COUNTRY (424), RIR (5), DNSSEC_ALGORITHM (8), TLD_OPERATOR (737), REGISTRAR (51K), EMAIL (237M), PHONE (60M), REGISTERED_PREFIX (326K, virtual), ANNOUNCED_PREFIX (1.4M, virtual), FEED_SOURCE (40, virtual), CATEGORY (18, virtual). All nodes have a "name" property. Threat-listed IPV4/IPV6/HOSTNAME nodes also carry: threatScore (Double), threatLevel (NONE/INFO/LOW/MEDIUM/HIGH/CRITICAL), threatSources, threatFirstSeen/threatLastSeen (epoch ms), and 13 boolean flags: isThreat, isAnonymizer, isC2, isMalware, isPhishing, isSpam, isBruteforce, isScanner, isBlacklist, isTor, isProxy, isVpn, isWhitelist. ANNOUNCED_PREFIX adds BGP-enrichment: isMoas, isAnycast, isWithdrawn, wasMoas, hasOriginChanged, threatScore, threatLevel, threatSourceCount, firstSeen, lastSeen. LISTED_IN edges carry firstSeen, lastSeen, weight.

KEY EDGES: RESOLVES_TO (HOSTNAME→IPV4/IPV6, forward only), CHILD_OF (child→parent: HOSTNAME→HOSTNAME→TLD), ALIAS_OF (CNAME), NAMESERVER_FOR / MAIL_FOR (NS/MX → domain - to list a domain's MX use (domain)<-[:MAIL_FOR]-(mx)), SPF_INCLUDE/SPF_IP/SPF_A/SPF_MX/SPF_EXISTS/SPF_REDIRECT (SPF policy; SPF_IP targets IPV4|IPV6|PREFIX), LINKS_TO (web hyperlinks, 10.8B), BELONGS_TO (3 semantics: IPV4/IPV6→PREFIX, PREFIX→RIR, FEED_SOURCE→CATEGORY), LOCATED_IN (IPV4/IPV6→CITY only - for country, chain through HAS_COUNTRY), HAS_COUNTRY (ASN/CITY/IPV4/HOSTNAME/PHONE/ANNOUNCED_PREFIX/REGISTERED_PREFIX→COUNTRY), ANNOUNCED_BY (IPV4/IPV6→ANNOUNCED_PREFIX, then ROUTES→ASN), ROUTES (ASN/ANNOUNCED_PREFIX→PREFIX/ASN, virtual), PEERS_WITH (ASN↔ASN, bidirectional, virtual), HAS_NAME (ASN→ASN_NAME, virtual; asn.name is the AS number - the network name lives on the ASN_NAME node), REGISTERED_BY (HOSTNAME/ASN/PREFIX→ORGANIZATION), HAS_REGISTRAR / PREV_REGISTRAR / HAS_EMAIL / HAS_PHONE (WHOIS), LISTED_IN (indicator→feed; threat intel for IPV4/IPV6/HOSTNAME), CONFLICTS_WITH (PREFIX/ANNOUNCED_PREFIX↔ASN, MOAS, bidirectional), OPERATES (TLD_OPERATOR→TLD).

TRAVERSAL CHAINS: DNS: HOSTNAME→RESOLVES_TO→IPV4→BELONGS_TO→PREFIX←ROUTES←ASN→HAS_NAME→ASN_NAME BGP-direct: IPV4→ANNOUNCED_BY→ANNOUNCED_PREFIX→ROUTES→ASN GeoIP: HOSTNAME→RESOLVES_TO→IPV4→LOCATED_IN→CITY→HAS_COUNTRY→COUNTRY WHOIS: HOSTNAME→HAS_REGISTRAR→REGISTRAR, HOSTNAME→HAS_EMAIL→EMAIL Threat: IPV4/HOSTNAME→LISTED_IN→FEED_SOURCE→BELONGS_TO→CATEGORY

RULES:

  • Use {name: "value"} or WHERE n.name = "value" for lookups - both indexed

  • Always include LIMIT on exploration queries (max 500)

  • shortestPath requires bounded depth: [*1..6]

  • Never scan FEED_SOURCE or CATEGORY directly - access via LISTED_IN from anchored nodes

  • STARTS WITH, ENDS WITH ".x", CONTAINS on .name are all indexed and fast

  • SIGNED_WITH currently returns 0 rows on live data (DNSSEC layer empty)

PROCEDURES: CALL explain("indicator") for threat assessment, CALL whisper.history("indicator") for historical WHOIS/BGP data, CALL whisper.variants("name") for typosquatting / brand-protection variant generation, CALL whisper.quota() for rate limits, CALL db.labels() / db.relationshipTypes() / db.schema("json") for schema introspection.

EXAMPLES: MATCH (h:HOSTNAME {name: "www.google.com"})-[:RESOLVES_TO]->(ip:IPV4) RETURN h.name, ip.name MATCH (ip:IPV4 {name: "8.8.8.8"})<-[:RESOLVES_TO]-(h:HOSTNAME) RETURN h.name LIMIT 20 MATCH (h:HOSTNAME {name: "google.com"})-[:RESOLVES_TO]->(ip:IPV4)-[:LOCATED_IN]->(c:CITY) RETURN ip.name, c.name MATCH (a:ASN {name: "AS15169"})-[:HAS_NAME]->(n:ASN_NAME) RETURN n.name MATCH (h:HOSTNAME) WHERE h.name ENDS WITH ".google.com" RETURN h.name LIMIT 20 MATCH (ip:IPV4 {name: "185.220.101.1"})-[:LISTED_IN]->(f:FEED_SOURCE) RETURN f.name, ip.threatScore

DOCUMENTATION: API reference: https://www.whisper.security/docs/cypher-api-reference Cypher query guide: https://www.whisper.security/docs/cypher-query-guide Cypher functions: https://www.whisper.security/docs/cypher-functions

Inputs: cypher (required). Returns: rows, error, columns, success, +4 more

run_recipeRun a WhisperGraph Catalog RecipeA
Read-onlyIdempotent
Inspect

Run a named whisper.security catalog recipe by its slug (see list_recipes for the full set). This is the highest-leverage tool for infrastructure & threat questions: instead of hand-writing Cypher, run the curated recipe.

Arguments:

  • recipe (string, required) - the recipe slug, e.g. "assess", "identify", "indicator-enrichment", "infrastructure-mapping", "attack-path", "subdomain-takeover", "typosquat", "bgp-hijack-exposure".

  • inputs (object, optional) - the recipe's inputs keyed by name (see the recipe's inputs[] in list_recipes). Examples: {"v":"8.8.8.8"} for assess/identify; {"value":"github.com"} for indicator-enrichment / infrastructure-mapping; {"domain":"paypal.com"} for typosquat; {"country":"BR"} for anycast-dns-root-sovereignty; {"value":"paypal.com","other":"paypa1.com"} for attack-path. Omit to use the recipe's built-in example.

  • params (object, optional) - flow tuning params, e.g. {"level":"deep"} for attack-surface / infrastructure-mapping / attack-path, {"depth":3} for blast-radius, {"instanceType":"Global"} for anycast-dns-root-sovereignty.

Behaviour:

  • direct recipes run keyless and return { success, recipe, mode:"direct", columns[], rows[], statistics }.

  • flow recipes need an API key (WHISPER_API_KEY over stdio, or the relayed X-API-Key / Authorization header over HTTP) and return { success, recipe, mode:"flow", steps[], totalLatencyMs } where each step carries { id, title, columns, rows }.

  • On an unknown slug or a keyless call to a keyed flow, returns { success:false, error, suggestion } - never throws.

Prefer run_recipe over hand-written Cypher whenever a recipe fits the question; fall back to the query tool for bespoke traversals.

Inputs: recipe (required), inputs, params. Returns: mode, rows, error, steps, +6 more

whisper_historyHistorical WHOIS / BGP for an IndicatorA
Read-onlyIdempotent
Inspect

Retrieve historical WHOIS or BGP data for a single indicator. The indicator can be an IPv4, IPv6, hostname, CIDR, or ASN - the procedure auto-detects the type.

Returns shape varies by indicator type:

  • IP / prefix (type=routing): { origin, prefix, startTime, endTime, peersSeing }

  • Domain (type=domain): WHOIS snapshots - { queryTime, createDate, updateDate, expiryDate, registrar, nameServers }

  • ASN (type=asn): prefix announcement history (slow, ~9s for large ASNs)

On upstream failure (the data source is rate-limiting or temporarily down), the row shape is: { available: false, error: "timeout" | ..., retryAfter: }. Surface the retryAfter to the user - DO NOT loop on retry.

Argument: indicator (string, required). Allowed characters: letters, digits, '.', '-', ':', '/', '_'.

Inputs: indicator (required). Returns: rows

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.6/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: schema inspection, typosquatting, threat assessment, label listing, recipe listing, general querying, recipe execution, and historical data retrieval. No overlap or ambiguity.

    Naming Consistency3/5

    Most tools use a verb_noun pattern (describe_label, list_labels, run_recipe), but domain_variants and whisper_history start with nouns, breaking consistency. Overall pattern is still readable but not uniform.

    Tool Count5/5

    8 tools is well-scoped for a graph intelligence server, covering essential operations without bloat or deficiency.

    Completeness5/5

    The tool set covers schema exploration, query execution, domain-specific analysis (typosquatting, threat assessment, historical data), and curated recipe execution. No obvious gaps for the stated purpose of infrastructure and threat intelligence analysis.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      D
      maintenance
      A Model Context Protocol tool that provides DNS querying capabilities for various record types (A, AAAA, MX, TXT, CNAME, NS, etc.) through a standardized MCP interface.
      1
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      A Model Context Protocol (MCP) server that provides comprehensive domain name research tools, including RDAP, WHOIS, and DNS query functionality.
      1
      2
      Apache 2.0
    • F
      license
      Not graded
      quality
      D
      maintenance
      Provides threat intelligence and vulnerability research tools by integrating with NVD, VirusTotal, AbuseIPDB, Shodan, and MITRE ATT\&CK. It enables users to perform CVE lookups, analyze IP reputation, and retrieve detailed MITRE ATT\&CK technique information.
      1
      -
    • A
      license
      A
      quality
      D
      maintenance
      Provides global internet resource intelligence by querying RIRs for IP and ASN data, routing visibility, and network health. It enables users to perform RPKI validation, BGP inspection, and historical allocation analysis through natural language or a REST API.
      42
      1
      MIT