MCP Office

Local-first, governed MCP servers for Microsoft Office — built for developers who want to treat Office files the way they treat code.

What is this?

MCP Office is a suite of Model Context Protocol servers that expose Microsoft Office capabilities as governed, deterministic tool calls. Each server runs locally on Windows, connects to your MCP client (Claude Desktop, VS Code Copilot, or any MCP-compatible client), and gives you structured control over Office files — without an in-app AI assistant.

This is not a Copilot replacement. It is a developer-first execution layer for Office automation.

Packages

✅ Available now

🚧 Coming next

MailMCP is now included in the repository's single mcp-office source distribution, but the currently published PyPI artifact does not yet claim MailMCP availability. See mailmcp/README.md for source validation while Windows Outlook UAT and release verification remain open.

New packages are added as they complete their proof cycle. See ROADMAP.md.

Quick start

Prerequisites

• Windows 10 or 11
• Python 3.11 or later (python --version)
• Claude Desktop or VS Code with GitHub Copilot
• Microsoft Office (Excel / PowerPoint / Word) — required for COM-backed tools (styling, PDF export, tracked-changes)
• Classic Microsoft Outlook desktop — required for MailMCP

Install

PyPI note: excelmcp on PyPI is an unrelated third-party package. Do not pip install excelmcp — that ships you a stranger's code. The published suite package is mcp-office.

pip install mcp-office

Or clone and install the consolidated suite from source (including MailMCP and Windows COM support):

git clone https://github.com/dosev-ai/mcp-office.git
cd mcp-office
python -m venv .venv
.venv\Scripts\activate
python -m pip install -e ".[com]"

Configure Claude Desktop

Open %APPDATA%\Claude\claude_desktop_config.json (create it if it doesn't exist) and add the servers you installed:

{
  "mcpServers": {
    "excel-excelmcp": {
      "command": "python",
      "args": ["-m", "excelmcp.server"],
      "env": {
        "EXCEL_ALLOWLIST_ROOTS": "C:\\path\\to\\your\\files",
        "EXCEL_ENABLE_WRITE": "true"
      }
    },
    "powerpoint-pptmcp": {
      "command": "python",
      "args": ["-m", "pptmcp.server"],
      "env": {
        "PPT_ALLOWLIST_ROOTS": "C:\\path\\to\\your\\files",
        "PPT_ENABLE_WRITE": "true"
      }
    },
    "word-wordmcp": {
      "command": "python",
      "args": ["-m", "wordmcp.server"],
      "env": {
        "WORD_ALLOWLIST_ROOTS": "C:\\path\\to\\your\\files",
        "WORD_ENABLE_WRITE": "true"
      }
    }
  }
}

Replace C:\\path\\to\\your\\files with the directory where your Office files live. Restart Claude Desktop after saving.

Verify

In Claude Desktop, send:

Call capabilities() on excel-excelmcp
Call capabilities() on powerpoint-pptmcp
Call capabilities() on word-wordmcp

Each should return a tool list (65 for Excel, 51 for PowerPoint, 50 for Word). If a server is missing, check that python resolves to the venv where you installed mcp-office.

Full per-package guides: excelmcp/README.md · pptmcp/README.md · wordmcp/README.md · mailmcp/README.md (source-integrated; release pending)

Detailed step-by-step: docs/quickstart.md

Requirements

Architecture

Your MCP client (Claude Desktop / VS Code Copilot / other)
        │
        │  MCP stdio protocol
        ↓
  MCP Office servers (local Python processes)
   ├─ excelmcp    — Excel automation (live)
   ├─ pptmcp      — PowerPoint automation (live)
   ├─ wordmcp     — Word automation (live)
   └─ mailmcp     — Outlook automation (source-integrated; release pending)
        │
        │  COM / Outlook COM / openpyxl / python-pptx / python-docx
        ↓
  Microsoft Office (local installation)

Each server is a standalone local stdio MCP process. MCP Office does not add a hosted service; Microsoft Office and your MCP client/model may still have their own sync and data-handling boundaries.

Contributing

This project is in active development. The best way to contribute right now:

1. Try any package (excelmcp, pptmcp, wordmcp) and open a First Run Report
2. Report bugs via GitHub Issues
3. Ask questions or share what you built in GitHub Discussions

See CONTRIBUTING.md for development guidelines.

License

MIT — see LICENSE.