cyber-mcp-tools
It lets you query MITRE ATT&CK and D3FEND data through ten MCP tools.
attack_search – Find techniques by text, platform, or tactic (compact summaries only).
attack_get – Fetch full details for a single technique by ID, optionally including subtechniques.
attack_detection – Get v19 detection strategies, optionally with analytics (log sources, tunable fields).
attack_mitigations – Get mitigations with technique-specific guidance.
attack_related – Walk the graph: actors/software that use the technique, subtechniques, or parent.
attack_actor – Look up groups, malware, tools, or campaigns by ID or name.
attack_meta – Get dataset version, counts, and valid tactic/platform filter values.
defend_countermeasures – Get D3FEND countermeasures for a technique, with sub-technique roll-up.
defend_search – Browse D3FEND countermeasures by defensive tactic or reverse-lookup affected techniques.
defend_meta – Check D3FEND coverage and which domains lack mappings.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@cyber-mcp-tools What detections and mitigations exist for T1055?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Cyber MCP Tools
An MCP server over MITRE ATT&CK and MITRE D3FEND — the attack side and the defence side of the same graph, joined on technique id, built so an agent can query them without dragging either dataset into its context window.
918 techniques enterprise 697 · mobile 124 · ICS 97
918 detection strategies with 2,053 analytics
2,017 mitigations each with technique-specific guidance
1,251 actors groups, malware, tools, campaigns
369 techniques with D3FEND countermeasures (154 distinct)ATT&CK v19.2. Ten tools. No network calls at runtime.
Official sources — MITRE ATT&CK · MITRE D3FEND · ATT&CK STIX data · D3FEND ontology
Quickstart
git clone https://github.com/FlemingJohn/cyber-mcp-tools
cd cyber-mcp-tools
npm install
npm run build:data # derives data/ from the MITRE sources
npm run buildThen point a client at it:
{
"mcpServers": {
"cyber-mcp-tools": {
"command": "node",
"args": ["/absolute/path/to/cyber-mcp-tools/dist/index.js"]
}
}
}Or, if deployed over HTTP, skip all of the above:
claude mcp add --transport http cyber-mcp-tools https://cyber-mcp-tools.vercel.app/mcpLive at https://cyber-mcp-tools.vercel.app
Per-client setup for Claude Code, Claude Desktop, Cursor and Antigravity is in INSTALL.md.
Related MCP server: MITRE ATT&CK MCP Server
The tools
Tool | Does | Typical cost |
| Find techniques by text, platform or tactic | ~130 tok |
| One technique by id | ~390 tok |
| Detection strategy, optionally with analytics | ~30 / ~340 tok |
| Mitigations with technique-specific guidance | ~200 tok |
| Paginated graph walk: usedBy, subtechniques, parent | ~380 tok |
| Group, malware, tool or campaign lookup | ~450 tok |
| Version, counts, valid filter values | ~35 tok |
| D3FEND countermeasures, with parent roll-up | ~73 tok |
| Browse by defensive tactic, or reverse lookup | ~55 tok |
| Coverage, and which domains have none | ~44 tok |
A full triage — search, read, detection, mitigations — costs about 2.1k tokens.
Why it is built this way
Loading ATT&CK with everything attached is roughly 1.46 million tokens. Every design decision here exists to make that impossible rather than merely discouraged.
Search returns summaries, never descriptions. Twenty results cost ~645 tokens. The same twenty as full objects cost ~17,130. Same navigational value, 26x cheaper.
Fan-out is paginated. T1105 is used by 520 actors, about 34,201 tokens if inlined.
attack_related returns 25 and tells you the total, for ~379 tokens.
The expensive shape is never the default. defend_countermeasures on T1055 is 73
tokens; detail: true is 4,477. attack_detection is ~30 tokens without analytics and
~340 with.
Ten tools, not eighty. Tool schemas sit in context on every turn, so breadth lives in parameters rather than in new tools.
Two things most ATT&CK tooling gets wrong
x_mitre_detection no longer exists. ATT&CK v19 removed the prose detection field —
zero of the 697 live Enterprise techniques carry it. Detection moved into first-class
objects reached through a chain:
technique ←detects─ detection-strategy → analytic → log sources + tunable fieldsTooling written before v19 still reads the old field and silently returns nothing. This
server walks the chain, so attack_detection returns the exact log channels to query and
the parameters that need tuning per environment.
D3FEND maps at sub-technique level, inconsistently. T1055 Process Injection has no
direct countermeasure, but ten of its children do. Ask a naive implementation and it
reports "no defences exist" — confidently, and wrongly. defend_countermeasures always
reports both the direct hits and which sub-techniques are covered.
What it cannot do
It never sees your environment. No logs, no telemetry, no deployed controls. It is a reference brain, not a detector. Anything it suggests needs validating against your real systems before anyone acts on it.
Analytics are detection ideas, not queries. They are not SPL, KQL or Sigma, and their tunable fields are deliberately unfilled because only you know your baseline.
D3FEND covers 40% of ATT&CK and none of mobile. 549 of 918 techniques have no countermeasure mapped. That is a gap in the mapping, not evidence a technique is undefendable — an important distinction, and one the tools are careful about.
The data is a snapshot. ATT&CK v19.2, frozen at build time. Refreshing means pulling the sources and rebuilding.
MITRE's own disclaimer applies and is reproduced in NOTICE: ATT&CK does not enumerate every technique, and covering it does not guarantee defensive coverage.
Where the data comes from
ATT&CK (attack.mitre.org) is read from a local clone of
mitre-attack/attack-stix-data — three
STIX 2.1 bundles. Set ATTACK_STIX_DATA to point elsewhere; it defaults to
../attack-stix-data.
D3FEND (d3fend.mitre.org) comes from the published inference output of the d3fend-ontology, vendored into the repo. The ontology itself asserts no direct ATT&CK links — the defence-to-attack edge is inferred by a reasoner over shared digital artifacts, so only the published inference result carries them.
Both builds are offline. Nothing is fetched at build or request time.
Development
npm run build:attack # rebuild data/attack from the STIX bundles
npm run build:defend # rebuild data/defend from the D3FEND mappings
npm run build:data # both
npm run build # compile to dist/
npm run dev # run from source over stdio
npm run typecheck # src and api
npm run inspect # 23 cases through the MCP Inspector CLI
npm run inspect:ui # the Inspector web clientnpm run inspect drives the server through the reference MCP client and exits non-zero on
failure, so it works as a CI gate. It checks that tools respond, that filters filter, that
unknown ids fail cleanly, that known-empty cases return empty rather than erroring, and
that no tool exceeds its token budget.
Architecture notes and the traps worth knowing before changing anything are in AGENTS.md. Worked call orders for triage, detection engineering, threat intel and gap analysis are in SKILLS.md.
License
The code is MIT — see LICENSE.
Everything under data/ is derived from MITRE ATT&CK and MITRE D3FEND. Both permit
redistribution, including commercially, and both require their notices travel with any
copy. Those notices are in NOTICE and must stay with any redistribution or
deployment.
ATT&CK and D3FEND are registered trademarks of The MITRE Corporation. This project is not affiliated with, endorsed by, or sponsored by MITRE.
Available Tools
10 toolsattack_actorLook up a threat group, malware or toolB
Find a group, malware, tool or campaign by ATT&CK id or name. Spans all domains, since many actors operate in more than one.
Inputs: idOrName (required), techniques.
attack_detectionGet detection strategy for a techniqueA
Return the ATT&CK v19 detection strategy for a technique. With analytics, includes log sources and tunable fields.
Inputs: id (required), analytics.
attack_getGet one ATT&CK techniqueA
Fetch a single technique by ATT&CK id such as T1055 or T1055.011. Ids are unique across domains.
Inputs: id (required), includeSubtechniques.
attack_metaATT&CK dataset version and valid filter valuesA
Return the dataset version and counts, plus the valid tactic slugs and platform names for filtering.
Inputs: section.
attack_mitigationsGet mitigations for a techniqueA
Return mitigations for a technique, including how each one applies to that specific technique.
Inputs: id (required), includeGeneral.
attack_searchSearch ATT&CK techniquesB
Find MITRE ATT&CK techniques by text, platform or tactic. Returns compact summaries only.
Inputs: query (required), limit, domain, tactic, +1 more.
defend_countermeasuresGet D3FEND countermeasures for a techniqueA
Return MITRE D3FEND countermeasures that counter an ATT&CK technique. D3FEND maps mostly at sub-technique level, so a parent id reports which of its children are covered. Pass detail for the full artifact-level payload. Mobile techniques have no D3FEND coverage.
Inputs: id (required), detail.
defend_metaD3FEND dataset coverage and limitsA
Return how much of ATT&CK the D3FEND mapping actually covers, including which domains have no coverage.
Inputs: no named parameters.
defend_searchBrowse D3FEND countermeasuresA
List D3FEND countermeasures, optionally filtered by defensive tactic, or find which ATT&CK techniques a named countermeasure applies to.
Inputs: limit, tactic, countermeasure.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
The attack_* and defend_* prefixes cleanly separate the ATT&CK and D3FEND domains, and search/get/actor/related have clear targets. The only mild ambiguity is among attack_detection, attack_mitigations, and defend_countermeasures, since all three describe defensive response content, though each names a distinct framework or strategy type.
Names consistently use lowercase snake_case and the attack_/defend_ prefixes, which is a recognizable pattern. However, the second part mixes nouns (actor, meta, mitigations), verbs (get, search), and adjectives (related), so it does not follow a single verb_noun convention.
Ten tools is a well-scoped size for a MITRE ATT&CK/D3FEND knowledge server. Each tool covers a distinct retrieval need without feeling padded or redundant.
The surface covers technique lookup and search, actor lookup, relationship traversal, detections, mitigations, D3FEND countermeasures, and metadata for both datasets. This provides a comprehensive read-only workflow for threat-intel queries with no obvious dead ends.
Related MCP Connectors
Query and retrieve information about various adversarial tactics and techniques used in cyber atta…
Knowledge graph for AI agents. Query concepts, walk edges, get advisories.
Query your OpsLevel internal developer portal: catalog, maturity data, and tech docs.
Enrich, search, assess, and manage threat intelligence through 80+ typed MCP tools.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables querying the MITRE ATT\&CK framework for adversarial tactics, techniques, mitigations, and detection methods through natural language, supporting both ID-based and fuzzy name-based searches.3-
- FlicenseBqualityDmaintenanceProvides comprehensive access to the MITRE ATT\&CK knowledge base with 50+ tools for querying threat actors, malware, and techniques, including automatic ATT\&CK Navigator layer generation for threat analysis and visualization.5545-
- AlicenseAqualityBmaintenanceEnables AI-native access to the MITRE ATT\&CK framework, allowing LLMs and agents to query techniques, threat groups, software, and generate ATT\&CK Navigator layers for threat intelligence and security workflows.65573 npm5Apache 2.0
- AlicenseAqualityAmaintenanceExposes the MITRE ATT\&CK framework to LLMs and AI assistants via the Model Context Protocol, enabling querying of techniques, tactics, groups, software, and mitigations.915MIT