feat(mcp): add clean_office tool (part of #61)

[avanithumballi]

What does this change?

Adds the first "no canvas" engine from #61: a clean_office MCP tool that strips author, editor, company, title and embedded thumbnail from .docx / .xlsx / .pptx files, running entirely on disk with no network.

Part of #61 (does not close it — the issue asks for several engines, one PR each).

Type of change

• 🧰 New tool or feature

How did you test it?

• New unit test in tests/mcp.test.ts cleans the report.docx fixture, re-opens the written file with unzipSync, and checks the author is gone from docProps/core.xml while word/document.xml is still present. Runs with npx vitest run tests/mcp.test.ts (11/11 pass).
• Tried the tool end to end via the MCP client: npx vitest run tests/mcp.test.ts exercises clean_office over the real InMemoryTransport, not just the engine.
• Full suite: npx tsc --noEmit clean; npx vitest run 214 passed / 1 failed (transcribe.test.ts, Windows CRLF line endings, fails on main too); npm run build:mcp passes; npm run build passes with no Node code in the browser bundle; npx playwright test 1139 passed / 1 failed (app.e2e.ts:378, pre-existing Windows touch-emulation issue, not in scope).

Also in this PR: a Windows fixture-path fix. new URL(..., import.meta.url).pathname returns /C:/Users/... on Windows, which Node resolves to C:\C:\Users\.... Two helpers switched to fileURLToPath. Without this, tests/mcp.test.ts couldn't run on Windows at all, and 26 e2e tests failed. This is a separate commit.

Checklist

• Nothing leaves the visitor's device (no new network requests, CDNs or third-party scripts)
• npm run check passes — see notes below
• npm run test:e2e passes — 1139/1140, one pre-existing Windows touch-emulation failure
• New user-facing text is added to src/i18n.ts (English) — N/A, this tool only runs on the MCP server, not in the browser UI
• New dependencies (if any) are permissively licensed and listed in THIRD_PARTY_NOTICES.md — no new dependencies: uses node:buffer and fflate, both already in the project

Notes on npm run check: I don't have that exact script name in my local checkout; I ran the equivalent commands the issue asked for — npx tsc --noEmit, npx vitest run, npm run build:mcp, npm run build, npx playwright test. Happy to run npm run check if it exists and I've missed it.

[avanithumballi]

Hi @kingrishabdugar — opened #93 for clean_office, the first "no canvas" engine you suggested. Node adapter, MCP tool, and a test that opens the output and checks its contents. Also included a small Windows fixture-path fix (fileURLToPath) that was blocking the test suite on Windows.

Next on my list is one of the Excel/CSV converters in src/engine/convert.ts — happy to start once this one lands, or in parallel if you prefer.

[kingrishabdugar]

Thanks a lot, @avanithumballi, and no worries at all about the timing, I hope the exams went well!

This is a really nice first engine: it reuses cleanOffice from the website as-is, the test opens the saved file and checks the author is gone while the content is still there, and the fileURLToPath change is a good catch for Windows. I ran it locally: typecheck, the full test suite and the MCP build all pass. I've approved the CI run, so the checks are running now.

Two small things before I merge:

1. mcp/README.md still says Office files are "coming". Could you add clean_office to its list of tools and update that line?
2. mcp/src/server.ts, mcp/src/tools.ts and tests/mcp.test.ts are missing the newline at the end of the file. Most editors have a setting for this.

Once those are in and CI is green, I'll merge it. And yes, an Excel/CSV converter from src/engine/convert.ts is a great next step. Same pattern, one tool per PR.

[avanithumballi]

Done — both changes are in:

• mcp/README.md: added clean_office to the tools table, updated the intro to mention Office files, and removed "Office files" from the "coming" line at the bottom.
• Trailing newlines added to mcp/src/server.ts, mcp/src/tools.ts and tests/mcp.test.ts.

Pushed as b73a832. CI should re-run automatically.

[avanithumballi]

@kingrishabdugar check this once

[kingrishabdugar]

Merged, thank you @avanithumballi! 🎉 clean_office is now part of the MCP server, and the Windows test fix will help everyone who works on Fizzdoc from Windows.

Whenever you're ready, the Excel/CSV converter from src/engine/convert.ts would be a great next one. Same pattern as this PR. No rush at all.