
Mitre Attack Mcp Server
io.github.imouichePublished by GitHub account imouichev0.1.6Updated Nov 14, 2026
MCP server providing 50+ tools for MITRE ATT&CK techniques, groups, and mitigations
Overview
Local MCP server that gives an assistant read-only query access to the MITRE ATT&CK knowledge base: techniques, groups, software, campaigns and mitigations.
- What you can do
- Look up and search techniques, tactics, sub-techniques and revoked techniques by ID, keyword, platform or tactic
- Explore threat groups, their aliases, the techniques and software they use, and campaign attribution
- Query malware and tools, campaigns, mitigations, data sources and data components, plus ICS assets
- Traverse relationships such as which groups use a technique, what mitigates what, and procedure examples
- Access any STIX object by STIX or ATT&CK ID, with created/modified filters and ID-to-name resolution
- Generate ATT&CK Navigator layers and report data status, entity counts and matrices
- When to use it
- Useful for threat-intelligence research, mapping detections and coverage gaps, profiling a threat actor's TTPs, or answering questions about ATT&CK techniques and mitigations across Enterprise, Mobile and ICS domains.
- Requirements
- Runs locally over stdio. Python 3.12 or newer on PATH; Node.js 16 or newer only for the npm path. First run downloads about 64MB of ATT&CK STIX data and installs dependencies into a private virtualenv under the home directory (roughly 290MB disk, about 490MB memory). Optional environment variables MITRE_MCP_PYTHON, MITRE_MCP_RUNTIME_DIR and MITRE_MCP_DATA_DIR relocate the interpreter and caches.
Installation
In SourceWeft
- Open Mitre Attack Mcp Server in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
MITRE ATT&CK MCP Server
65 MCP tools giving LLMs and agentic systems machine-callable access to the MITRE ATT&CK knowledge base
[MCP tools] [Domains] [ATT&CK] [npm] [npm downloads] [License] [MCP Registry] [MCPRush security rating]
Installation · Quick Start · Security · Data Cache · Tools · Examples
Overview
The MITRE ATT&CK MCP Server exposes the MITRE ATT&CK framework over the Model Context Protocol, giving LLMs and agentic systems structured, machine-callable access to adversary tactics, techniques, threat groups, software, campaigns and mitigations.
It reads the official STIX 2.1 bundles published by MITRE through the
mitreattack-python library. No scraping, no third-party API, no
reverse-engineered endpoints.
65 tools across three domains
Covering Enterprise, Mobile and ICS, with relationship traversal throughout — which groups use a technique, which software a campaign deployed, what mitigates what — plus ATT&CK Navigator layer generation for coverage and gap visualisation. All output is shaped for LLM consumption.
Registered in the official MCP Registry as
io.github.imouiche/mitre-attack-mcp-server.
Installation
Python (PyPI)
Or with uv:
Node (npm)
Or without installing:
The npm package needs a few minutes on first run. It ships the Python source, not the Python dependencies. On first launch it creates a private virtualenv under
~/.mitre-mcp-server/runtime/, installs the pinned dependencies into it, and downloads the ATT&CK STIX data. Subsequent runs start in a few seconds.Some MCP clients time out waiting for a server to respond, so warm the cache once from a terminal before configuring your client:
Python 3.12 or newer must be on
PATH.uvis used when available andvenvpluspipotherwise. Nothing is installed into your system or currently active Python environment.
Environment variables
From source
Quick Start
1. Configure your client
Add the server to claude_desktop_config.json:
- macOS —
~/Library/Application Support/Claude/claude_desktop_config.json - Windows —
%APPDATA%\Claude\claude_desktop_config.json
2. Restart the client
Quit Claude Desktop completely (Cmd+Q on macOS) and reopen it.
3. Query
"What techniques does APT29 use for initial access?"
"Generate an ATT&CK Navigator layer for ransomware groups"
"Show me all Windows persistence techniques"
Security
MCPRush rates this server security A with low access scope and no sign-in required. That assessment is MCPRush's own, not an independent audit, but it reflects properties you can verify directly in this repository:
- No credentials. The server reads no API keys, tokens or secrets. The only environment variables it consults are the three cache-location overrides documented above.
- Narrow network scope. The sole outbound request is the ATT&CK STIX
download, performed by MITRE's own
mitreattack-pythonlibrary. There are no telemetry, analytics or third-party calls. - Integrity-checked data. Downloads are verified against the expected hash published in MITRE's release manifest, so a corrupted or substituted bundle is rejected rather than loaded.
- Local execution. The server runs as a local stdio process. It opens no listening port and exposes no HTTP surface.
- Read-only behaviour. All 65 tools are queries over the cached STIX data. Nothing mutates state outside the cache directory, and no tool executes shell commands or evaluates caller-supplied code.
Found a security issue? Please open an issue.
Data Cache
The server does not bundle ATT&CK data. On first start it downloads the official STIX 2.1 bundles for all three domains and caches them on disk, then loads them into memory. Later starts reuse the cache and skip the download.
Location — ~/.mitre-mcp-server/data/v{version}/, overridable with
MITRE_MCP_DATA_DIR.
Current release — ATT&CK v20.1, about 64MB across three files:
Clearing the cache to force a re-download
The cache is never invalidated automatically. If the data is stale, partially downloaded, or corrupted, clear it and the server will fetch a fresh copy the next time it starts.
Inspect what is currently cached:
Re-download the current release, replacing what is there:
Or simply delete the cache and let the server re-fetch on next start. This is
the most reliable method and the one to use with the npm install, which has no
convenient python on PATH:
To refresh a single domain rather than all three:
Reclaiming disk space from superseded releases
When MITRE publishes a new ATT&CK version, the server downloads it into a new
v{version} directory and leaves the previous one in place. Over time
these accumulate — a machine that has run through two releases will hold both:
Only the directory matching the current mitreattack-python release is read.
Delete the others:
The npm install path keeps a second cache, the virtualenv under
~/.mitre-mcp-server/runtime/. It is keyed by package and Python version, so
upgrades create a new directory and leave the old one behind. Removing the
whole runtime directory is safe; it is rebuilt on the next launch.
Available Tools
65 tools covering the major ATT&CK entities and their relationships. All tool
handlers are asynchronous. Most accept an optional domain argument of
enterprise, mobile or ics, defaulting to enterprise.
Infrastructure and metadata
Generic STIX access
Lower-level accessors that work across every object type, useful when you hold a STIX ID rather than an ATT&CK ID.
Techniques
Groups
Software
Campaigns
Mitigations
Tactics, data sources and ICS
Example Queries
ATT&CK Navigator Visualisation
The generate_layer tool produces ATT&CK Navigator-compatible JSON.
- Ask your client to generate a layer, for example "Generate an ATT&CK Navigator layer for all techniques used by APT29"
- Save the JSON output to a file, for example
apt29_layer.json - Upload it to ATT&CK Navigator
Typical uses include mapping red-team exercise coverage, highlighting unmonitored techniques, profiling a threat actor's TTPs, and showing which techniques are mitigated against which are exposed.
Worked example with LangGraph
- Article — From MITRE ATT&CK to agentic threat investigation, on building a multi-agent LangGraph system over these tools
- Live demo — Interactive Gradio demo on Hugging Face Spaces
Technical Details
Architecture
- Language — Python 3.12 or newer
- Protocol — Model Context Protocol via FastMCP, served over stdio
- Data source — official STIX 2.1 bundles through
mitreattack-python - Concurrency — asynchronous tool handlers
- Caching — all three domains loaded into memory at startup, so queries need no disk access
Data
Counts track the pinned ATT&CK release and change when MITRE publishes a new one. See Data Cache for refreshing.
Requirements
- Python — 3.12 or newer
- Node.js — 16 or newer, for the npm installation path only
- Disk — about 64MB of ATT&CK data plus about 200MB of Python
dependencies. The npm path keeps its dependencies in its own virtualenv, so
budget roughly 290MB under
~/.mitre-mcp-server/. - Memory — about 490MB resident with all three domains loaded, since every domain is held in memory for fast queries. Measured on Python 3.12 against ATT&CK v20.1.
mcp— pinned to>=1.1.0,<2. The server uses the v1mcp.server.fastmcpAPI, which mcp 2.0 renamed toMCPServer. An unpinned install resolves 2.x and fails at import.
License
Apache License 2.0. See LICENSE.
Issues and pull requests are welcome at github.com/imouiche/complete-mitre-attack-mcp-server.
Author
Inoussa Mouiche, Ph.D. — University of Windsor, WASP Lab. Research in threat intelligence automation, machine learning, multi-agent security systems and LLM-powered security operations.
GitHub · LinkedIn · Google Scholar · [email protected]
Acknowledgments
Built on MITRE ATT&CK using mitreattack-python, MITRE's official library, and implementing the Model Context Protocol. Thanks to the MCP developer community, including R. Jasper.
MITRE ATT&CK is a registered trademark of The MITRE Corporation. This project is not affiliated with or endorsed by The MITRE Corporation.
Source: README.md at commit e5278c9
Tools
0Version history
2- v0.1.6LatestNov 14, 2026
- v0.1.1Sep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts