dataharpoon

DataHarpoon is an MCP-ready query engine that connects to your data sources and lets you query them via natural language and raw SQL. It exposes an MCP server interface so you can issue federation-style queries through MCP clients, while also supporting direct SQL-style queries when needed. The server is designed to work with datasets defined in files or registered data sources, enabling you to join, filter, and aggregate data with familiar SQL constructs or by asking natural-language questions that MCP translates into SQL behind the scenes. You can also run the included example calls via the call_mcp integration demonstrated in the README, which shows how to classify issues or perform other tasks by delegating to external tools.

Prerequisites:

• Rust toolchain (Rust and Cargo)
• Git

Install and run locally:

1. Clone the repository:

git clone https://github.com/mrasu/dataharpoon.git
cd dataharpoon

2. Build the project:

cargo build

3. Run the binary directly (for local testing):

# After building, the binary will be at ./target/debug/dataharpoon
./target/debug/dataharpoon serve mcp -c /path/to/data_harpoon.toml

4. Create an MCP server configuration (example):

{
  "mcpServers": {
    "dataharpoon": {
      "command": "/path/to/dataharpoon",
      "args": [
        "serve",
        "mcp",
        "-c",
        "/path/to/data_harpoon.toml"
      ]
    }
  }
}

5. Run as an MCP server using the configuration above:

# If you have a runner or orchestrator, point it to the config file.

Prerequisites summary:

• Rust/Cargo installed
• Git installed
• A data_harpoon.toml configuration file defining data sources and schemas

Tips and considerations:

• Ensure the data_harpoon.toml correctly references your data sources (files or databases) and defines the expected schema so MCP can translate natural language prompts into accurate SQL.
• When integrating with external tools (as shown in the README example with call_mcp and Claude), ensure any required API keys or credentials are supplied via environment variables or secure configuration files.
• The MCP server can be started with the command shown in the mcp_config; adjust paths to your actual binary and config file.
• If you run into port or networking issues, verify firewall rules and that the MCP client can reach the server endpoint defined by your orchestration layer.
• Log locally for troubleshooting; enable verbose logs if available in your build to capture the generated SQL and any translation steps.