Skip to main content
Model Context Protocol (MCP) lets your model call tools you expose via an MCP server.

​
Single MCP server

Define an input schema for your tool: 
type CalculatorInput struct {
  Operation string  `json:"operation" jsonschema_description:"The arithmetic operation to perform (add, subtract, multiply, divide)"`
  X         float64 `json:"x" jsonschema_description:"First number"`
  Y         float64 `json:"y" jsonschema_description:"Second number"`
}
Initialize KarmaAI with MCP options: 
kai := ai.NewKarmaAI(
  ai.GPT4o,         // use a model that supports MCP/tool calling
  ai.OpenAI,        // pick a provider
  ai.WithMaxTokens(1000),
  ai.WithTemperature(1),
  ai.WithTopP(0.9),
  ai.WithTopK(50),

  // MCP config
  ai.SetMCPUrl("http://localhost:8086/mcp"),
  ai.SetMCPAuthToken(config.GetEnvRaw("TEST_TOKEN")),
  ai.SetMCPTools(ai.MCPTool{
    FriendlyName: "Calculator",
    ToolName:     "calculate",
    Description:  "Perform basic arithmetic operations (add, subtract, multiply, divide).",
    InputSchema:  CalculatorInput{},
  }),
)
Ask the model to use your tool: 
history := models.AIChatHistory{
  Messages: []models.AIMessage{
    {
      Role:    models.User,
      Message: "Please calculate 123 + 456. Use the calculator tool provided.",
    },
  },
  ChatId:    "example-chat-2",
  CreatedAt: time.Now(),
  Title:     "Using MCP Tools",
}

resp, err := kai.ChatCompletion(history)
if err != nil { panic(err) }
fmt.Println(resp.AIResponse)
Ensure your MCP server is running and reachable at the URL you configured, and that the auth token matches.

​
Multiple MCP servers

Use either SetMCPServers to replace servers or AddMCPServer to append. 
// Replace any existing MCP servers
func SetMCPServers(servers ...MCPServer) ai.Option

// Append a new MCP server
func AddMCPServer(url, authToken string, tools ...MCPTool) ai.Option
Example usage: 
kai := ai.NewKarmaAI(
	ai.GPT4o,         // use a model that supports MCP/tool calling
  ai.OpenAI,        // pick a provider
  ai.AddMCPServer("http://localhost:8086/mcp", "token1",
    ai.MCPTool{FriendlyName: "Calc", ToolName: "calculate", InputSchema: CalculatorInput{}},
  ),
  ai.AddMCPServer("https://tools.mycompany.com/mcp", "token2",
    ai.MCPTool{FriendlyName: "Weather", ToolName: "weather_lookup", InputSchema: struct {
      City string `json:"city"`
    }{}},
  ),
)
Note: Not all models support MCP. Use a model where Model.SupportsMCP() is true (e.g., Grok variants shown above).