Show and tell: MCP spreadsheet tool server for WorkPaper agents

[gregkonush]

@bilig/headless@0.11.34 now has a dependency-free MCP-style WorkPaper tool-server example.

The shape is intentionally small:

• tools/list exposes read_workpaper_summary and set_workpaper_input_cell with JSON Schema inputs.
• tools/call returns text content plus structuredContent with before/after computed values, edited address, previous value, new value, formula persistence, and restored readback checks.
• The example runs locally with no hosted service, no API key, and no agent framework dependency.

Run it from a clean checkout:

git clone https://github.com/proompteng/bilig.git
cd bilig/examples/headless-workpaper
npm install
npm run agent:mcp-tools

Links:

• Guide: https://proompteng.github.io/bilig/mcp-workpaper-tool-server.html
• Example script: https://github.com/proompteng/bilig/blob/main/examples/headless-workpaper/mcp-tool-server.mjs
• Package: https://www.npmjs.com/package/@bilig/headless

What I want feedback on:

1. Should the next example be stdio transport, HTTP/SSE, or a concrete MCP SDK adapter?
2. Which spreadsheet workflow should the write tool prove next: pricing, forecast, budget approval, quote rules, or something else?
3. What structured fields would you expect every spreadsheet write tool to return before trusting it in an agent loop?

If this saves you a spreadsheet-tooling spike, star the repo so the next person searching for MCP spreadsheet tools can find it: https://github.com/proompteng/bilig/stargazers

[gregkonush]

Update from the MCP/server-directory pass:

• The packaged stdio server is now installable with npm exec --package @bilig/headless -- bilig-workpaper-mcp.
• Official MCP Registry entry is live as io.github.proompteng/bilig-workpaper: https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.proompteng%2Fbilig-workpaper
• Glama listing is live: https://glama.ai/mcp/servers/proompteng/bilig
• Claude Desktop MCPB bundle guide is live: https://proompteng.github.io/bilig/claude-desktop-mcpb-workpaper.html
• Directory/status page is live: https://proompteng.github.io/bilig/mcp-spreadsheet-server-directory.html

The part I want feedback on now: if you use MCP clients for spreadsheet or finance workflows, which setup should be the shortest copy-paste path next: Claude Desktop, Cursor, VS Code, Codex, Cline, Zed, Continue, or something else?

I am especially interested in whether the write tool's structured readback is enough to trust an agent edit. The useful path should be: read workbook summary, set one input cell, recalculate formulas, return before/after values, and persist the WorkPaper JSON without a hosted account.

[gregkonush]

Update: the official MCP Registry entry is current again for @bilig/headless@0.14.0.

Verified today:

• Registry API returns 9 Bilig WorkPaper entries and marks 0.14.0 as latest.
• The published npm package still installs with npm exec --package @bilig/headless -- bilig-workpaper-mcp.
• A rebuilt Claude Desktop MCPB bundle packages bilig-workpaper@0.14.0.
• The generated stdio server reports serverInfo.version: "0.14.0" and exposes read_workpaper_summary plus set_workpaper_input_cell.

Directory/status page: https://proompteng.github.io/bilig/mcp-spreadsheet-server-directory.html
MCP client setup: https://proompteng.github.io/bilig/mcp-client-setup.html

Feedback I want from MCP users: is the first useful write tool still "set one input cell and return readback", or should the server prove a higher-level workflow next, such as setting named inputs, validating ranges, or saving restored WorkPaper JSON through a storage adapter?

[gregkonush]

Update: the official MCP Registry entry is current for @bilig/headless@0.14.2.

Verified on May 14, 2026:

• Registry API returns 10 Bilig WorkPaper entries.
• 0.14.2 is marked latest for io.github.proompteng/bilig-workpaper.
• npm latest is @bilig/headless@0.14.2.
• Install command is unchanged: npm exec --package @bilig/headless -- bilig-workpaper-mcp.

Registry publish run: https://github.com/proompteng/bilig/actions/runs/25850256208
Directory/status page: https://proompteng.github.io/bilig/mcp-spreadsheet-server-directory.html

The feedback ask is the same: for agent workflows, is the single-cell write plus formula readback enough, or should the next tool be named inputs, range validation, or storage-backed save/restore proof?

[gregkonush]

Registry refresh: io.github.proompteng/bilig-workpaper is now published as @bilig/headless@0.14.5 in the official MCP Registry.

Verification:

curl -fsSL 'https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.proompteng%2Fbilig-workpaper'

The live API returns 11 matching Bilig WorkPaper entries and marks 0.14.5 as isLatest: true. The publish workflow run is here: https://github.com/proompteng/bilig/actions/runs/25868014475

The install command is still:

npm exec --package @bilig/headless -- bilig-workpaper-mcp

Useful feedback at this point: should the next MCP proof focus on named input schemas, range-level validation errors, or storage-backed save/restore across multiple tool calls?

[gregkonush]

Update: the official MCP Registry entry is refreshed to @bilig/headless@0.14.9.

Verification:

• Publish workflow: https://github.com/proompteng/bilig/actions/runs/25882165865
• Live Registry query: https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.proompteng%2Fbilig-workpaper
• Current latest entry: io.github.proompteng/bilig-workpaper with npm package @bilig/headless@0.14.9, transport.type: stdio, and isLatest: true.

The install command remains:

npm exec --package @bilig/headless -- bilig-workpaper-mcp

For MCP users evaluating workbook tools, the next useful feedback is still concrete: should the server expose named input schemas, stricter range-level validation errors, or storage-backed save/restore across multiple tool calls first?