Configure MCP Server

AttackLens includes a Model Context Protocol (MCP) server that enables AI assistants like Claude to interact with your security data in real time. Once configured, you can use natural language to query assets, findings, vulnerabilities, attack paths, and compliance status without leaving your AI workflow.

INFO

Requires Admin role or higher to configure the MCP server. Users with Viewer role or higher can use the MCP server through their AI assistant once it is enabled.

What Is the MCP Server?

The Model Context Protocol (MCP) is an open standard that allows AI assistants to connect to external data sources and tools. The AttackLens MCP server exposes your security data through this protocol, giving AI assistants structured access to:

• Assets: Query asset inventory, metadata, linked sensors, and group memberships.
• Findings: Search and filter security findings by severity, policy, asset, and status.
• Vulnerabilities: Look up vulnerability details, CVSS scores, EPSS probabilities, CISA KEV status, and affected assets.
• Attack Graph: Query attack paths, chokepoints, toxic combinations, and node relationships.
• Policies and Rulesets: Review policy configurations, rule definitions, and compliance status.
• Posture: Get posture scores and compliance summaries across frameworks.

The MCP server runs as part of your AttackLens deployment. All data stays on your infrastructure -- nothing is sent to external AI providers by the MCP server itself. The AI assistant connects to the MCP server directly and processes the data in its own context.

Step 1: Enable the MCP Server

Navigate to Integrations > MCP Server.

Toggle the Enable MCP Server switch to turn on the server.

Step 2: View Connection Details

After enabling, the page displays the connection details your AI assistant needs:

Field	Description
Server URL	The MCP server endpoint (e.g., https://your-attacklens-instance:8080/mcp).
Transport	The connection transport method (typically sse for server-sent events).
Authentication	API key or token required to connect.

WARNING

The API key controls access to your security data. Treat it like any other credential: do not share it in unencrypted channels, do not commit it to version control, and rotate it periodically.

Step 3: Configure Access Scope

The access scope determines what data the MCP server exposes to connected AI assistants. You can restrict access to specific data categories:

Scope	Data Accessible
Assets	Asset list, details, inventory, groups, conflicts.
Findings	Security findings, posture evaluation results.
Vulnerabilities	Vulnerability findings, CVE details, EPSS/KEV data.
Attack Graph	Attack graph nodes, edges, paths, chokepoints, toxic combinations.
Policies	Policy definitions, rulesets, compliance frameworks.
Discovery	Adapter connections, discovery snapshots, cloud resources.
Settings (Admin only)	System configuration (read-only).

By default, all scopes are enabled. Disable scopes you do not want the AI assistant to access.

TIP

For most deployments, enabling all scopes provides the best experience. The AI assistant can provide more comprehensive analysis when it has access to the full security picture. Restrict scopes only if you have specific data isolation requirements.

Step 4: Connect Your AI Assistant

Claude Desktop

To connect Claude Desktop to your AttackLens MCP server:

1. Open Claude Desktop settings.
2. Navigate to the MCP Servers section.
3. Click Add Server.
4. Enter the server configuration:

{
  "mcpServers": {
    "attacklens": {
      "url": "https://your-attacklens-instance:8080/mcp",
      "transport": "sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

5. Save and restart Claude Desktop.

Other MCP-Compatible Clients

Any MCP-compatible AI client can connect using the same server URL, transport, and authentication details. Refer to your client's documentation for the specific configuration format.

What Can the AI Assistant Do?

Once connected, you can interact with your security data using natural language. Example queries:

Asset Analysis

• "Show me all assets running Windows Server 2019 or older."
• "Which assets have critical vulnerabilities and are internet-facing?"
• "List all assets in the Production Servers group with their sensor status."

Vulnerability Analysis

• "What are the top 10 vulnerabilities by EPSS probability across my environment?"
• "Are any of my assets affected by CVE-2024-3094?"
• "Show me all CISA KEV vulnerabilities that are not yet remediated."

Attack Path Analysis

• "What are the shortest attack paths to our database servers?"
• "Identify the chokepoints that appear in the most attack paths."
• "Which toxic combinations exist in our environment?"

Compliance and Posture

• "What is our current compliance status against ISO 27001?"
• "List all critical findings from the CIS benchmark evaluation."
• "Which policies have the lowest pass rate?"

Remediation Guidance

• "What should we prioritize fixing first based on attack graph analysis?"
• "Generate a remediation plan for the critical vulnerabilities on server-prod-01."

Security Considerations

WARNING

The MCP server provides read-only access to your security data. It cannot modify assets, change configurations, or perform any write operations. However, the data it exposes may be sensitive. Consider the following:

• Network access: The MCP server is accessible only from within your network (or VPN). It does not need to be exposed to the internet.
• API key rotation: Rotate the API key periodically, especially if a team member who had access leaves the organization.
• Audit: MCP server access is logged in the AttackLens audit log. Review access patterns periodically.
• AI data handling: Understand how your AI assistant processes data. When using Claude, your data is processed according to Anthropic's data policies. For air-gapped environments, consider using a locally-hosted AI model.

Disabling the MCP Server

To disable the MCP server:

1. Navigate to Integrations > MCP Server.
2. Toggle the Enable MCP Server switch off.
3. Click Save.

All active MCP connections will be terminated immediately. Connected AI assistants will no longer be able to query your data.