Show and tell: verifying workbook formulas through MCPStdioPlugin

[gregkonush]

I put together a small Semantic Kernel Python example for a local workbook MCP server because I kept running into the same pattern: an agent needs spreadsheet-style business logic, but driving Excel or a browser just to edit inputs and read calculated outputs is a brittle dependency.

The example imports a local MCP server with MCPStdioPlugin, edits an input cell, reads the recalculated formula result, persists the workbook JSON, and then reopens it to verify the readback. No Excel UI, no browser automation, no cached XLSX formula value assumptions.

Docs and proof page:
https://proompteng.github.io/bilig/semantic-kernel-workpaper-mcp.html

Source:
https://github.com/proompteng/bilig/tree/main/examples/semantic-kernel-workpaper-mcp

The local run is intentionally boring:

uv run --python 3.12 --with 'semantic-kernel[mcp]' \
  python examples/semantic-kernel-workpaper-mcp/semantic_kernel_workpaper_mcp.py \
  --local-source

Expected proof includes:

{
  "verified": true,
  "pluginName": "BiligWorkPaper",
  "loadedTools": [
    "get_cell_display_value",
    "list_sheets",
    "read_cell",
    "set_cell_contents",
    "validate_formula"
  ]
}

This is not a request to add anything to Semantic Kernel itself. I am sharing it because the MCPStdioPlugin path is a clean fit for deterministic workbook tools, especially when agents need a reviewable calculation model rather than a spreadsheet UI session.

If there is a better SK-native shape for this kind of example, I am happy to adjust the sample around that pattern.