Hugging Face WorkPaper MCP Space

Use this when an agent already works with Hugging Face Spaces, Gradio, or MCP tools and needs a small WorkPaper readback fixture before wiring a private workbook or service.

Live Space: https://huggingface.co/spaces/gregkonush/bilig-workpaper-mcp-readback

The live Space runs the Docker template from this repo. The Gradio function has MCP enabled, edits Inputs!B3, recalculates dependent formulas, exports WorkPaper JSON, restores it, and returns the exact values it read back.

Source in this repo:

examples/huggingface-workpaper-space

Run Locally

cd examples/huggingface-workpaper-space
npm install --omit=dev --package-lock=false
uv run --python 3.12 --with 'gradio[mcp]>=6.0,<7' python app.py --check

The check calls the same Node script the Space uses and expects:

{
  "editedCell": "Inputs!B3",
  "verified": true
}

What The Tool Returns

The response includes:

• edited cell: Inputs!B3;
• formulas for Summary!B2:B5;
• before, after, and restored readback values;
• serialized WorkPaper JSON byte count;
• checks for formula output changes, restored readback, input persistence, and formula persistence.

For the default win rate 0.4, Summary!B3 reads back as 96000.

Deploy Your Own

Create a Hugging Face Docker Space and upload the files from examples/huggingface-workpaper-space. The public Space is a no-key hosted fixture for agent stacks that want to inspect a workbook API result without spreadsheet UI. It is a discovery and evaluation artifact, not user-file storage.

For private files, run the file-backed WorkPaper MCP server:

npm exec --package @bilig/workpaper@latest -- \
  bilig-workpaper-mcp \
  --workpaper ./pricing.workpaper.json \
  --init-demo-workpaper \
  --writable