MCP Node

Overview

The MCP Node enables seamless integration with external data sources, APIs, and tools through the Model Context Protocol (MCP). This node allows AI models to securely access external resources, retrieve real-time data, and execute specific functions in a structured manner. It's particularly useful for applications requiring dynamic data access, real-time information retrieval, and integration with third-party services.

Features

What can I build?

1. Weather Information Systems - Retrieve real-time weather data for specific locations
2. Financial Data Applications - Access stock prices, currency conversion rates, and market data
3. Document Processing Systems - Search and retrieve documents from various sources
4. API Integration Workflows - Connect to third-party services and APIs
5. Real-time Monitoring Systems - Access live data feeds for monitoring and alerts
6. Custom Tool Integration - Use specialized tools and functions provided by MCP servers

Setup

Configure MCP Server

1. Go to Lamatic Studio
2. Navigate to MCP/Tools > MCP
3. Click on + Add MCP
4. Configure the MCP server with connection details:
   • Credential Name: Enter a descriptive name for your MCP server
   • Host: Provide the MCP server URL (required field)
   • Type: Select the protocol type (http or sse)
   • Headers: Configure authentication and custom headers
5. Click "Test & Save" to validate the connection

Select the MCP Node

1. Add the MCP node to your flow
2. Select the configured MCP server
3. Choose the specific function or tool to execute
4. Configure input parameters for the selected function
5. Deploy the project

Configuration Reference

Low-Code Example

nodes:
  - nodeId: mcpNode_195
    nodeType: mcpNode
    nodeName: MCP
    values:
      memories: "[]"
      messages: "[]"
      attachments: ""
      credentials: MCP Server
      selectedTools:
        - add_comment_to_pending_review
        - add_issue_comment
        - assign_copilot_to_issue
        - cancel_workflow_run
        - create_and_submit_pull_request_review
      promptTemplate: Tell me details about github user ABC
      generativeModelName:
        type: generator/text
        model_name: default
        credentialId: Credential
        provider_name: openai
        credential_name: AI_API
    modes: {}
    needs:
      - triggerNode_1

Output

{
  "_meta": {
    "prompt_tokens_cost": 0.07718000000000001,
    "completion_tokens_cost": 0.0027,
    "total_cost": 0.07988,
    "cached_prompt_tokens_cost": 0,
    "prompt_tokens": 7718,
    "completion_tokens": 90,
    "total_tokens": 7808,
    "prompt_tokens_details": {
      "cached_tokens": 0,
      "audio_tokens": 0
    },
    "completion_tokens_details": {
      "reasoning_tokens": 0,
      "audio_tokens": 0,
      "accepted_prediction_tokens": 0,
      "rejected_prediction_tokens": 0
    },
    "model_name": "gpt-4-turbo-preview",
    "model_provider": "openai",
    "info": {
      "max_tokens": 4096,
      "max_input_tokens": 128000,
      "max_output_tokens": 4096,
      "input_cost_per_token": 0.00001,
      "output_cost_per_token": 0.00003,
      "supports_pdf_input": true,
      "supports_function_calling": true,
      "supports_parallel_function_calling": true,
      "supports_prompt_caching": true,
      "supports_system_messages": true,
      "supports_tool_choice": true
    }
  },
  "generatedResponse": "response"
}

_meta

• A nested object containing metadata about the MCP function execution and model usage.

  • prompt_tokens_cost: Cost of prompt tokens used in the request.
  • completion_tokens_cost: Cost of completion tokens generated.
  • total_cost: Total cost of the entire operation.
  • cached_prompt_tokens_cost: Cost of cached prompt tokens (if any).
  • prompt_tokens: Number of prompt tokens used.
  • completion_tokens: Number of completion tokens generated.
  • total_tokens: Total number of tokens used.
  • prompt_tokens_details: Detailed breakdown of prompt token usage.
    • cached_tokens: Number of cached tokens used.
    • audio_tokens: Number of audio tokens (if applicable).
  • completion_tokens_details: Detailed breakdown of completion token usage.
    • reasoning_tokens: Number of reasoning tokens used.
    • audio_tokens: Number of audio tokens generated.
    • accepted_prediction_tokens: Number of accepted prediction tokens.
    • rejected_prediction_tokens: Number of rejected prediction tokens.
  • model_name: Name of the AI model used (e.g., "gpt-4-turbo-preview").
  • model_provider: Provider of the AI model (e.g., "openai").
  • info: Model capabilities and configuration information.
    • max_tokens: Maximum tokens allowed for output.
    • max_input_tokens: Maximum tokens allowed for input.
    • max_output_tokens: Maximum tokens allowed for output.
    • input_cost_per_token: Cost per input token.
    • output_cost_per_token: Cost per output token.
    • supports_pdf_input: Whether the model supports PDF input.
    • supports_function_calling: Whether the model supports function calling.
    • supports_parallel_function_calling: Whether the model supports parallel function calling.
    • supports_prompt_caching: Whether the model supports prompt caching.
    • supports_system_messages: Whether the model supports system messages.
    • supports_tool_choice: Whether the model supports tool choice.

generatedResponse

• The response generated by the MCP node, which includes the result of the MCP function execution processed through the AI model.

Troubleshooting

Common Issues

Debugging

1. Check Lamatic Flow logs for error details.
2. Verify MCP server configuration and connectivity.
3. Test the MCP server connection independently.
4. Review function documentation for correct parameter usage.