The Office Worker — office-worker-mcp

The complete document MCP for AI agents. Generate, edit, secure, audit and extract professional PDF, Word, Excel and PowerPoint documents — fully local, zero API keys, deterministic and safe. Built for Hermes, Claude and Cursor via MCP.

Requirements (read this first)

pip install office-worker-mcp          # core: MCP server + CLI 'owi'

Connect it to your agent (stdio):

{ "mcpServers": { "office-worker": { "command": "office-worker-mcp" } } }

Full setup — all 30 tools, zero friction (recommended)

The core pip package covers the full document lifecycle; three capabilities need system binaries that pip cannot ship (LibreOffice for Office→PDF conversion, Tesseract for OCR, Pango/Cairo native libs for WeasyPrint). The official Docker image bundles everything and runs on any OS with Docker — including Windows, no WSL2 needed:

docker pull ghcr.io/cafra-house/office-worker-mcp:latest

{
  "mcpServers": {
    "office-worker": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-v", "${HOME}/office-worker-data:/data", "ghcr.io/cafra-house/office-worker-mcp:latest"]
    }
  }
}

Prefer bare metal? Run owi doctor after pip install — it reports exactly which capability is inactive and prints the one-line install command for your OS (apt/brew/dnf/winget). Details in docs/docker.md and docs/extras.md.

See it in action

Real documents generated by the MCP itself — no manual editing, no cloud:

• Excel — create_excel structured table + autofilter + edit_excel add_chart / add_pivot native pivot sheet
• Word — create_word executive report with headings, bullets and a data table, corporate theme
• PPTX — create_pptx multi-slide deck with a native DrawingML bar chart
• PDF — render_document(design_mode="premium") editorial layout with metrics table

What you get

One package covers the entire document lifecycle — create → read → extract → convert → secure → sign → audit — with honest fidelity reporting (rich | clean | lossy) and proactive next_steps guidance on every creation tool.

• Create — Word / Excel / PPTX / PDF from templates or declarative blocks; corporate themes, logos, watermarks, premium editorial design mode
• Read & extract — structured Markdown/JSON for RAG (read_office, pdf_extract_structured) + embedded images as base64 for multimodal agents
• Convert — PDF ↔ Office ↔ CSV bidirectionally with explicit fidelity warnings
• Secure — irreversible redaction (pdf_redact), metadata scrubbing (scrub_metadata), password encryption (protect_office), form flattening for archival compliance
• Sign & audit — real PAdES signatures (sign_pdf) + cryptographic verification (verify_pdf_signature), honest document diffing (document_diff)
• Productivity — mail merge (mail_merge), native Excel pivots (add_pivot), native PPTX charts, multi-chapter books with auto TOC (create_book)
• Onboarding — owi doctor environment diagnostics with per-OS install hints; packaged skills installable via owi skill install

Efficiency (measured)

~2550 tokens/turn across 30 specialized tools vs ~6400 for a typical 47-tool CRUD suite — less bloat, more stable agents, lower cost. Measured, not claimed.

Documentation

Develop & test

pip install -e ".[dev,book]"
pytest -q            # full suite verified on disk (46 passed)
python count_tokens.py   # live token-budget audit (2550.75 tok v0.9.0, <2600 target)

License

MIT — see LICENSE. Acknowledgements in NOTICE. Maintained by CaFra-House.