MCP Server

scitex-io provides a Model Context Protocol (MCP) server, enabling AI agents to save, load, and discover file formats autonomously.

Installation

pip install scitex-io[mcp]

Starting the Server

scitex-io mcp start

MCP Client Configuration

Add to your MCP client configuration (e.g., Claude Desktop, Cursor):

{
  "mcpServers": {
    "scitex-io": {
      "command": "scitex-io",
      "args": ["mcp", "start"]
    }
  }
}

Available Tools

**Table 3.** Four MCP tools for AI-assisted file I/O. All tools accept JSON parameters and return JSON results.
Tool	Description
`io_list_formats`	List all registered save/load format extensions
`io_load`	Load data from any supported file format. Returns type, shape, and preview.
`io_save`	Save data (as JSON string) to any supported format
`io_register_info`	Show how to register custom format handlers with examples

Tool Details

io_load

{
  "path": "/data/experiment.csv",
  "format": null,
  "cache": true
}

Returns shape/length, type name, and a truncated preview of the loaded data.

io_save

{
  "data_json": "{\"x\": [1, 2, 3], \"y\": [4, 5, 6]}",
  "path": "/data/output.json",
  "verbose": false
}

Diagnostics

scitex-io mcp doctor          # Check MCP dependencies
scitex-io mcp list-tools      # List available tools
scitex-io mcp list-tools -vv  # With full parameter descriptions