Skip to main content

MCP Tool Calling

Tool calling enables AI models to interact with external systems and execute functions beyond text generation. Headlamp-KAITO integrates with the Model Context Protocol (MCP) to provide powerful tool-calling capabilities directly within your Kubernetes environment.

Overview​

Tool calling allows AI models to:

  • Execute Functions: Call predefined functions with structured parameters
  • Access External APIs: Interact with external services and databases
  • Perform Actions: Execute system commands and operations
  • Retrieve Real-time Data: Access live data from various sources

Model Compatibility​

  • Note: Currently, only Llama models support tool calling in Headlamp-KAITO. The system provides visual indicators for tool-capable models 🔧:

Visual indicator for Model supporting tool calling

MCP Server Management​

Authentication Methods​

Headlamp-KAITO supports MCP servers using two authentication methods:

  • Authorization Header: Credentials are sent via HTTP headers for secure access.
  • URL Path Authentication: Credentials are included in the endpoint URL for server authentication.

Transport Layer​

Headlamp-KAITO supports MCP servers using both streamable HTTP and Server-Sent Events (SSE) protocols. This enables real-time communication and efficient data streaming between the AI model and external tools, ensuring responsive tool execution and seamless integration.

Adding MCP Servers

Users can add MCP servers through the management interface: Visual indicator for Model Chat tool calling MCP Server Management

  1. Server Name: Human-readable identifier
  2. Endpoint URL: HTTP endpoint for the MCP server
  3. Description: Optional description of server capabilities
  4. API Key: Optional authentication token
  5. Auth Method: Choose between header or URL authentication

Server Status Monitoring​

The interface shows real-time server status: MCP Server Status

  • Connection Status: Green indicator for connected servers
  • Tool Count: Number of tools available from each server
  • Error States: Visual indication of connection problems

Configuration Persistence​

MCP server configurations are stored in the component state and can be: Visual indicator for toggling MCP Servers

  • Enabled/Disabled: Toggle servers on/off
  • Edited: Modify server settings
  • Deleted: Remove unused servers

Usage Examples​

Github MCP Server Setup​

Fill in the following:

Server name: Github MCP Server (or whatever you'd like)
Endpoint URL: https://api.githubcopilot.com/mcp/
Description (optional): Official GitHub MCP server for repository management, issues, pull requests, and more
API Key: enter your Github personal access token
Authentication Method: Authorization Header

Then click Add Server

Github MCP Server setup

You're all set! Chat with your model with queries with related to the Github MCP server context. See the available tools for this server at the Github repository: https://github.com/github/github-mcp-server

Github MCP Server Conversation