Sigrid MCP server for Sigrid On-Premise

This documentation covers on-premise Sigrid. It is not applicable for cloud-based Sigrid.

The Sigrid MCP server lets AI coding assistants use Sigrid’s analysis while developers work. When using Sigrid On-Premise, the Sigrid MCP server runs inside your own cluster as an optional service. Unlike the other optional components, such as the LDAP group sync and the Open Source Health knowledge base updater, which are jobs, the MCP server is a long-running service with its own mcp subchart in sigrid-stack.

Prerequisites

• You should have already read the other Sigrid On-Premise documentation.
• Sigrid On-Premise is installed and working, using the sigrid-stack Helm chart version 1.0.20260917 or later.
• Your Sigrid license includes the Sigrid MCP server.

The MCP server uses the sigrid-multi-analyzer image, which was previoulsy only used in CI/CD for code analysis.

Enabling the MCP server

The MCP server is disabled by default. The minimal configuration to enable it is shown below. The MCP server validates Sigrid tokens with auth-api and retrieves data from sigrid-api, both via their cluster-internal service names. These are prefixed with your Helm release name, sigrid-onprem in this example. The server is exposed on the same host as Sigrid, under the /mcp path.

mcp:
  enabled: true
  env:
    # A list in your values file replaces the chart's default list, so the defaults are repeated here.
    - name: JAVA_TOOL_OPTIONS
      value: -Djava.security.egd=file:/dev/urandom -XX:MaxRAMPercentage=80.0 -Djava.io.tmpdir=/data
    - name: TMPDIR
      value: /data
    - name: TPF_PARALLEL_ANALYSIS
      value: "1"
    - name: MATOMO_BASE_URL
      value: "disabled"
    - name: JWKS_URI
      value: http://sigrid-onprem-auth-api/oauth2/jwks
    - name: SIGRID_API_BASE_URL
      value: http://sigrid-onprem-sigrid-api/api/v1
  ingress:
    className: "nginx"  # Use the same ingress controller as the other Sigrid services
    annotations: { }  # Specify any ingress controller-specific annotations you need
    hosts:
      - host: "my-sigrid.example.com"
        paths:
          - path: /mcp
    tls:
      - secretName: my-sigrid-tls
        hosts:
          - "my-sigrid.example.com"

The subchart requests 7 CPU and 28Gi memory per replica by default. Lower mcp.resources if your node pool cannot accommodate this, keeping in mind that the MCP server runs Sigrid analyses on code sent by your AI agent, so it needs memory comparable to an analysis job in your CI pipeline.

By default the subchart also creates a NetworkPolicy that limits egress to auth-api and sigrid-api, and a HorizontalPodAutoscaler that needs the Kubernetes metrics API. Set mcp.networkPolicy.enabled: false or mcp.autoscaling.enabled: false if your cluster does not support these.

Connecting an AI coding assistant

After the mcp pod is running, developers connect their AI coding assistant to https://my-sigrid.example.com/mcp using an authentication token from your on-premise Sigrid. The Sigrid MCP Integrations page lists the configuration per IDE.

Contact and support

Feel free to contact SIG’s support team for any questions or issues you may have after reading this documentation or when using Sigrid.