MITRE ATLAS Agent

July 31, 2026 · View on GitHub

Version 1.1.0 — Single-Agent Release · ATLAS data v2026.7

The ATLAS Agent is an open-source AI assistant for exploring the MITRE ATLAS Knowledge Base.

It gives you three ways to work with ATLAS:

  1. A natural-language guide for any user
    Learn about ATLAS tactics, techniques, mitigations, and case studies by asking questions in plain English.
  2. MCP server and API access
    Use the agent and its capabilities inside compatible tools, workflows, and custom applications.
  3. Fully open-source and customizable
    Inspect, modify, and tailor prompts, behavior, and tools to fit specific security workflows and research needs.

Ask questions. Explore relationships. Generate knowledge graphs. Integrate ATLAS into your workflow.

How It Works

The ATLAS Agent uses a single agent with direct access to three purpose-built tools. This streamlined architecture reduces handoffs and complexity for faster responses while retaining the core ATLAS research capabilities:

  • Vector Store: performs semantic search over ATLAS entities.
  • Dataset: performs structured lookup and relationship extraction from the ATLAS dataset.
  • Graph Visualizer: generates focused graph URLs from verified ATLAS IDs.

ATLAS Agent Architecture

The bundled dataset and vector store are synchronized with the MITRE ATLAS data release v2026.7.

Getting Started

The ATLAS Agent is built with Langflow and is designed to run with your preferred tool-compatible LLM provider.

Requirements:

  • uv

Step 1. Start Langflow

Navigate to atlas-agent/ and run the following command.

uv run langflow run --env-file .env

(Press CTRL+C to quit)

Step 2. Configure Models

Open the flow named ATLAS Agent v1.1.0.

On the highlighted model component, enter the following values for your LLM provider:

  • Model Name
  • API Base URL
  • API Key

Langflow offers several LLM components.

Step 3. Start Chatting

Open the Langflow playground and start chatting with the ATLAS Agent.

Example questions:

  • What ATLAS techniques are related to prompt injection?
  • What mitigations apply to model poisoning?
  • Show me case studies involving data poisoning.
  • Generate a graph for techniques related to AI supply chain compromise.

Access the ATLAS Agent as an MCP Server

Every Langflow project runs a separate MCP server that exposes the project's flows as MCP tools. See Langflow as MCP Server.

Release Notes

See CHANGELOG.md for release history and upgrade notes.

Notice

Copyright 2026 The MITRE Corporation. Approved for Public Release; Distribution Unlimited. Case Number 26-1336.