Skip to content

Model Context Protocol (MCP)

By enabling this toolset, HolmesGPT will be able to connect to external Model Context Protocol (MCP) servers, extending its capabilities with custom tools and data sources.

MCP allows you to integrate external tools and services that implement the MCP specification, providing a standardized way to extend HolmesGPT's functionality.

Prerequisites

  1. An external MCP server running and accessible
  2. MCP server endpoint URL
  3. Any required authentication credentials for the MCP server

Configuration

Add the following to ~/.holmes/config.yaml, creating the file if it doesn't exist:

toolsets:
  mcp/remote:
    enabled: true
    config:
      servers:
        - name: "custom-tools"
          endpoint: "http://your-mcp-server:8080"
          # Optional authentication
          headers:
            Authorization: "Bearer <your-token>"
        - name: "another-server"
          endpoint: "https://mcp.example.com"

holmes:
  toolsets:
    mcp/remote:
      enabled: true
      config:
        servers:
          - name: "custom-tools"
            endpoint: "http://your-mcp-server:8080"
            # Optional authentication
            headers:
              Authorization: "Bearer <your-token>"
          - name: "another-server"
            endpoint: "https://mcp.example.com"

Advanced Configuration

You can configure multiple MCP servers with different settings:

toolsets:
  mcp/remote:
    enabled: true
    config:
      timeout: 30  # Request timeout in seconds
      max_retries: 3  # Maximum number of retries
      servers:
        - name: "primary-tools"
          endpoint: "http://primary-mcp-server:8080"
          timeout: 60  # Server-specific timeout
          headers:
            Authorization: "Bearer <token>"
            X-API-Version: "v1"
        - name: "backup-tools"
          endpoint: "https://backup-mcp-server.com"
          verify_ssl: false  # For development/testing

MCP Server Examples

Popular MCP servers you can integrate:

  • Database MCP: Direct database query capabilities
  • Cloud Provider MCP: AWS/Azure/GCP specific tools
  • Monitoring MCP: Custom monitoring system integration
  • Internal Tools MCP: Company-specific tooling

Capabilities

The capabilities depend on the connected MCP servers. Common tool types include:

Tool Category Description
Data Sources Access to external databases, APIs, or files
Custom Commands Organization-specific operational commands
Specialized Analysis Domain-specific analysis tools
External Integrations Third-party service integrations

Dynamic Capabilities

The exact tools available through MCP servers are discovered dynamically when the toolset connects to each server.

Creating Your Own MCP Server

To create a custom MCP server for your organization:

  1. Implement the MCP specification
  2. Expose your tools via the MCP protocol
  3. Configure HolmesGPT to connect to your server
  4. Your custom tools will be automatically available to Holmes

See the MCP documentation for implementation details.