Meeting Notes Transcript Summary

Create .docx Word documents from Claude

Say "write a proposal for Beta Corp, checkout rebuild, 4,500 EUR, three phases" and get a real .docx you can send. This MCP server writes Word documents from chat, proposals, quotes, service agreements, statements of work and letters, with your letterhead, headings, bullet and numbered lists and tables. It also turns markdown into .docx, reads an existing .docx back as text and outline, and fills {{placeholders}} in a template you already use, keeping every style, table, header and image of the original. Everything runs locally: no upload, no account, no native dependency.

In the [official MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers/io.github.theluckystrike%2Fdocx-document-generator-proposal-contract-markdown/versions/latest) (io.github.theluckystrike/docx-document-generator-proposal-contract-markdown).

Real Word documents from chat, proposals, contracts and letters, without a template site or an office subscription.

60-second install

npm publish for @theluckystrike/mcp-docx is pending. Until then, the .mcpb one-click bundle or a clone+build is the working path, both are verified below.

One-click (.mcpb): download docx.mcpb from the latest release and double-click it in Claude Desktop: https://github.com/theluckystrike/mcp-servers/releases/latest

(claude_desktop_config.json):

{

"mcpServers": { "docx": { "command": "npx", "args": ["-y", "@theluckystrike/mcp-docx"] } } }

Claude Code:

claude mcp add docx -- npx -y @theluckystrike/mcp-docx

(.cursor/mcp.json):

{

"mcpServers": { "docx": { "command": "npx", "args": ["-y", "@theluckystrike/mcp-docx"] } } }

The npx form above starts working the moment the package is published. Until then, use the .mcpb bundle above, or build from source with exactly these three commands:

git clone https://github.com/theluckystrike/mcp-servers.git && cd mcp-servers

npm install npm run build -w packages/mcp-license -w servers/docx

Then point your client's command at node with one arg: the absolute path to servers/docx/dist/index.js.

To run in Pro mode set MCP_LICENSE_KEY in the same config block, or call license_activate once with your key.

Tools

Resource: docs://recent returns the last 25 documents written, newest first, with kind, client, reference and path. Prompt: write_proposal_from_hours turns tracked hours (or an invoice_summary from mcp-time-tracker) into a priced proposal.

There is no doc_to_pdf, and that is deliberate

Every pure-JavaScript path from Word to PDF needs either a native dependency (LibreOffice, a Chromium binary, a C++ rendering library) or a cloud API. This collection ships neither, so npx works on any machine with Node and nothing else. doc_to_html writes semantic HTML with a print stylesheet instead: open it, print to PDF, and the result is what you would have got. The tool description and the tool's own answer say the same thing, so the model does not promise a PDF it cannot produce.

What you can say

Worked example

You: Write a proposal for Beta Corp. Checkout rebuild, 4,500 EUR, 50% on

signature 50% on delivery, three phases: discovery 1 week, build 3 weeks, launch 1 week. Valid until the end of the year.

proposal_create { client: "Beta Corp", project_title: "Checkout rebuild", summary: "Beta Corp loses orders at checkout. This project rebuilds it.", scope: ["Audit the current funnel", "Rebuild the checkout", "Ship and measure"], deliverables: ["New checkout in production", "A one-page handover"], timeline: [{phase: "Discovery", duration: "1 week"}, {phase: "Build", duration: "3 weeks"}, {phase: "Launch", duration: "1 week"}], price: {amount: 4500, currency: "EUR", terms: "50% on signature, 50% on delivery"}, valid_until: "2026-12-31" } -> PROP-2026-0001, EUR 4,500.00 -> ~/.local/share/mcp-servers/docx/documents/checkout-rebuild.docx

The file opens in Word, Pages, LibreOffice and Google Docs: letterhead, cover title, "Prepared for Beta Corp", Summary, Scope of work, Deliverables, a Timeline table, an Investment table with EUR 4,500.00, the payment terms, and a signature block for both parties. Every amount carries its currency code, no bare numbers.

Free vs Pro

Pro is a one-time $19, or $39 for every server in the collection, lifetime.

How the .docx is produced and read

Documents are written with docx, a pure-JavaScript OOXML writer, no native module, no headless browser, no office install.

Reading and template filling use no dependency at all. A .docx is a ZIP, so node:zlib opens it and a small WordprocessingML walk pulls out paragraphs, heading levels (from w:pStyle), list items and tables in document order. Numbered lists are told apart from bullets by resolving each paragraph's w:numId against word/numbering.xml, which is the only place that distinction is recorded, without it every numbered list reads back as bullets.

Template filling substitutes on the joined text of each paragraph, not per run. Word routinely breaks a placeholder you typed as {{client}} into three runs ({{cli, ent}}, ...) after an edit or a spell-check pass, and per-run replacement silently misses those, the document comes back with the placeholder still in it. The replaced text goes into the first run, keeping its formatting, and the remaining runs of that paragraph are blanked. Every other part of the package is copied byte-for-byte, so styles, images, headers, footers and section setup survive. A placeholder with no value is left in place and reported, never blanked.

Existing files are never overwritten

Every tool that writes a file (doc_create, doc_from_markdown, doc_to_html, doc_fill_template, proposal_create, contract_create) refuses an out_path that already exists and tells you so:

Error: /path/acme-proposal.docx already exists and nothing was written.

Pass overwrite: true to replace it, or give a different out_path.

The path is reserved with an exclusive create, not an existence check, so two processes writing the same out_path at the same time cannot clobber each other: one wins, the other is refused and writes nothing.

When you do not pass an out_path, the file name is derived from the title, and a derived path never lands on an earlier document: a second proposal with the same title is written as ...-2.docx. To change a proposal you already sent, use proposal_update {reference} instead: it rewrites the same file from the stored structured data and keeps the reference number.

The check runs before anything is built or written, so a refused call leaves the disk untouched and burns no reference number. Pass overwrite: true when replacing the file is what you want.

Characters Word cannot carry

XML 1.0 allows TAB, LF and CR but no other control code. Every string that reaches word/document.xml is cleaned first: control codes and unpaired surrogates are removed, literal \n escapes become real paragraph breaks, and stray whitespace collapses. When something was removed the tool says so in its answer instead of handing you a file Word would offer to repair.

How it stores data

The business profile, the document register and the reference counter live under ${XDG_DATA_HOME:-~/.local/share}/mcp-servers/docx/ as plain JSON, plus a documents/ subfolder holding the generated files when you do not pass out_path. Every mutating call runs inside an advisory lock on .../docx/.lock, so two clients on one data directory cannot allocate the same reference number or lose a record. Saves go to a temporary file and are renamed into place.

If one of those JSON files is unreadable or not valid JSON, it is never treated as "empty". The file is moved aside byte-for-byte as .json.corrupt- , a .json.corrupt marker is written, and every tool returns data file is corrupt; moved to ...; nothing was written until you restore a good copy and delete the marker.

Reference numbers are PROP-YYYY-NNNN for proposals and AGR-YYYY-NNNN for agreements. The counter is written before the record is stored, so a crash burns a number rather than reusing one, and existing numbers are scanned so a restored register can never hand back a reference that is already on a sent document.

Limits and honest caveats

• No PDF output. Use doc_to_html and print. See the section above for why.
• doc_read reads `.d