MCP Servers

MCP (Model Context Protocol) Servers provide external integrations and capabilities for agents through a standardized protocol. They enable agents to interact with external systems, APIs, and services in a consistent way.

MCP Servers act as bridges between agents and external systems, providing tools and resources that agents can use during their execution. They follow the Model Context Protocol specification for consistent integration.

Example YAML

apiVersion: ark.mckinsey.com/v1alpha1
kind: MCPServer
metadata:
  name: github-mcp
  namespace: default
spec:
  address:
    valueFrom:
      serviceRef:
        name: github-mcp-server
        namespace: default
        port: http
        path: mcp
  transport: http
  description: "GitHub repository operations via MCP protocol"

Usage with Agents

MCP servers are accessed through Tool resources, which agents then reference:

MCPServer → Tool → Agent

# Agent references MCP tools
apiVersion: ark.mckinsey.com/v1alpha1
kind: Agent
metadata:
  name: github-agent
spec:
  prompt: You are a GitHub assistant with access to repository operations.
  modelRef:
    name: gpt-4-model
  tools:
    - type: custom
      name: github-get-repo  # References Tool that connects to MCP server

See Tools for creating Tool resources that connect to MCP servers.

Key Features

• Standardized Model Context Protocol implementation
• HTTP and stdio transport support
• Service reference integration with Kubernetes
• Secure credential management
• Tool and resource discovery

Sample Resources

See the samples/mcp directory for additional examples including:

• GitHub MCP server integration
• Local MCP server setup
• External MCP server configuration
• MCP server with authentication

Next: Learn about Tools for creating custom agent capabilities.