Netcdf

An MCP server that reads and understands NetCDF files, designed to give coding agents context about scientific data files.

Installation

uv pip install -e .

Usage

With Cursor/Claude

Add to your MCP configuration file (~/.cursor/mcp.json):

{
  "mcpServers": {
    "netcdf-mcp": {
      "command": "/path/to/uv",
      "args": ["run", "--directory", "/path/to/netcdf-mcp", "netcdf-mcp"]
    }
  }
}

Replace /path/to/uv with your uv binary location (e.g., /opt/homebrew/bin/uv) and /path/to/netcdf-mcp with where you cloned this repository.

Standalone

netcdf-mcp

Tools

Tool	Description
get_file_structure	Get dimensions, coordinates, variables, and attributes
list_variables	List all data variables with shapes and units
get_global_attributes	Get file-level metadata
get_variable_info	Detailed info about a specific variable
get_variable_stats	Statistical summary (min, max, mean, percentiles)
get_variable_sample	Extract sample values for inspection
get_dimension_values	Get coordinate values (time, lat, lon)
check_data_quality	Check for NaN, outliers, and suspicious patterns
compare_variables	Compare two variables for compatibility and correlation
get_time_range	Get temporal extent in human-readable format
get_spatial_bounds	Get geographic extent (lat/lon bounds)
check_cf_compliance	Verify CF conventions compliance
explain_variable	Human-readable explanation of a variable