Turn a chat into a real Word document: proposals, quotes, contracts, statements of work and letters as .docx files, plus reading and filling existing .docx templates. Everything stays on your machine.
mcp-docx has 10 tools with complete schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (business_set, doc_create, proposal_create). Descriptions are present but often lack actionable context, e.g., doc_create says 'Create a .docx from blocks' without explaining when to use it vs doc_from_markdown. Parameter descriptions are minimal (avg ~40 chars vs baseline 72). No output schemas documented. Error handling is basic (fail() returns generic text). Security is reasonable (no secrets in params, file operations validated). Composition is good, tools are single-purpose and chainable. Naming clarity is strong except 'doc_list' (should be 'docs_list' for consistency with list_ pattern).
Read your stored business details
Store the sender details printed on every proposal, contract and letter, plus the currency, VAT, terms and prefix defaults. It writes the SAME shared profile as the invoice server's business_set.
Create a contract .docx with automatic numbering and business letterhead
Create a .docx from blocks: headings, paragraphs, bullet and numbered lists, tables and code blocks
Fill {{placeholders}} in a .docx template and write the result
Create a .docx from Markdown: ATX headings, paragraphs, bullet and numbered lists, GFM pipe tables, fenced code blocks
List all stored documents
Output schemas not documented. Tools return structured responses (e.g., doc_read returns blocks, doc_list returns documents) but LLMs cannot see the field structure. This forces agents to guess what fields are available and risks downstream tool call failures.
Parameter descriptions are sparse. 'blocks' in doc_create and proposal_create lacks detail on structure (is it an array of objects? what fields?). 'values' in doc_fill_template says 'Placeholder values to fill' but doesn't explain the object shape or how placeholders are named.
Tool descriptions lack context for selection. 'doc_create' vs 'doc_from_markdown' both create .docx files but the description doesn't explain when to use each. LLMs must infer from names alone, risking wrong tool choice.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Read a .docx and return its blocks: headings, paragraphs, bullet and numbered lists, tables and code blocks
Convert a .docx to HTML
Create a proposal .docx with automatic numbering and business letterhead
Error handling is generic. The fail() helper returns 'Error: <text>' with no guidance on recovery. E.g., if doc_fill_template fails due to missing placeholder, the LLM gets no hint to call doc_read first to inspect the template structure.
Naming inconsistency: 'doc_list' should be 'docs_list' to match the list_ verb pattern baseline (list_users, list_channels). Current name is ambiguous, could mean 'list a single doc' or 'list all docs'.