Insight Digger

Insight Digger MCP provides an enterprise-grade Model Context Protocol system that connects your Claude Desktop or custom clients to a dedicated MCP server layer. It enables secure authentication, intelligent caching, multi-user session isolation, and seamless integration with your backend data APIs to drive guided, LLMed workflows.

How to use

You use the MCP by connecting your client (such as Claude Desktop or a custom REST client) to either the hosted MCP service or a local MCP server you run yourself. Start by choosing a connection method, then initialize a session, fetch available data sources, configure your analysis, and execute the analysis to view results and dashboards. The system manages sessions, security, and parameter caching to streamline multi-step workflows across users and tools.

How to install

Prerequisites: Node.js 18+ and Python 3.8+ are commonly used in the examples. You also need access to your Insight Digger backend API.

# Option 1: Claude Desktop Integration (recommended)

# Install the NPX bridge
npx @sandsiv/data-narrator-mcp

# If configuring Claude Desktop, add the MCP server entry (data-narrator) with the following settings in your Claude Desktop config:
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "data-narrator": {
      "command": "npx",
      "args": ["-y", "@sandsiv/data-narrator-mcp@1.0.0"],
      "env": {
        "MCP_CLIENT_URL": "https://your-mcp-service.com"
      }
    }
  }
}

# Option 2: Direct API Integration (for developers)

# Start the MCP Services
# Install dependencies
./scripts/setup/install_dependencies.sh

# Activate virtual environment
source venv/bin/activate

# Start the Flask API service
npm run dev:flask
# OR
python src/python/scripts/start_flask_api.py

Configuration and deployment notes

Environment must be secured with HTTPS in production. The MCP client URL and backend API URL are provided via environment variables or via startup arguments. Token lifetimes and session handling are designed to prevent cross-user data leakage while enabling seamless re-authentication when needed.

Security & production readiness

Key security aspects include immediate credential validation during initialization, strict session isolation, proper HTTP status codes for errors, and input validation for API URLs and JWT tokens. Tokens are not logged or stored permanently, and MCP instances are created only for valid credentials. Ensure all production communications use HTTPS and that enterprise authentication is integrated as needed.

Tools & workflow

The MCP provides a set of LLMed tools to manage data source discovery, analysis configuration, execution, and results presentation. These tools are designed to be invoked in sequence to guide end-to-end data analysis.

Troubleshooting and notes

If you encounter authentication errors, verify JWT validity and URL connectivity. For tool-related errors, review backend responses and ensure the session is active. Automatic cleanup and re-authentication prompts help recover from session issues.

Interacting with the MCP servers

Two connection methods are available: a hosted HTTP MCP server and local stdio MCP servers that run on your machine. The hosted HTTP method uses a remote URL, while the local stdio method starts a process on your host to provide MCP endpoints.

Available tools

list_sources

Discover available data sources that MCP can access.

get_source_structure

Retrieve the structure of a selected data source for deeper analysis.

prepare_analysis_configuration

Configure analysis parameters and options before execution.

execute_analysis_from_config

Run the prepared analysis configuration and generate results.