Word Live MCP Server
A Windows MCP server that attaches to a running Microsoft Word instance via COM automation and exposes 20 tools for reading, editing, formatting, and exporting documents — live on screen.
README
Word Live MCP Server
A Windows MCP (Model Context Protocol) server that attaches to a running Microsoft Word instance via COM automation and exposes 20 tools for reading, editing, formatting, and exporting documents — live on screen.
The Word counterpart to the PowerPoint Live MCP.
How It Works
The server attaches to the running Word.Application instance via win32com (pywin32). Because it talks to the live COM object, every edit appears instantly on the Word canvas — no file round-trip, no reopen.
Tools (20)
| Category | Tool | Description |
|---|---|---|
| Discovery | list_documents |
List all open Word documents |
get_active_document |
Details of active document (name, path, word/para count) | |
| Read | get_document_content |
Paginated paragraph listing with styles |
get_paragraph |
Full text + formatting of one paragraph | |
get_table |
All cell content from a table | |
get_selection |
Current cursor/selection state | |
find_text |
Full-text search with paragraph locations | |
| Edit | set_paragraph_text |
Replace text of a paragraph |
insert_paragraph_after |
Insert new paragraph after a given index | |
delete_paragraph |
Delete a paragraph | |
replace_text |
Find-and-replace throughout document | |
| Format | set_font |
Font size/bold/italic/color/name |
set_highlight |
Yellow/red/green/none highlight | |
set_paragraph_style |
Apply named style (Heading 1, Normal, etc.) | |
| Tables | set_table_cell |
Set cell text by (row, column) |
insert_table |
Insert a new table | |
| Export | export_pdf |
Export document to PDF |
export_page_image |
Export single page as PNG (via PDF render) | |
| Other | update_toc |
Update all Tables of Contents |
save_document |
Save the document |
Requirements
- Windows + Microsoft Word installed
- Python 3.11+ with these packages:
Runpip install pywin32 mcp pymupdfpython Scripts/pywin32_postinstall.py -installonce after installing pywin32.
Installation
Hermes Agent
Add to ~/.hermes/config.yaml:
mcp_servers:
word:
command: "C:/Users/You/AppData/Local/Programs/Python/Python313/python.exe"
args: ["C:/path/to/word-live-mcp/server.py"]
connect_timeout: 30
enabled: true
timeout: 90
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"word": {
"command": "python",
"args": ["C:/path/to/word-live-mcp/server.py"]
}
}
}
Generic MCP Client
The server communicates over stdio using the MCP protocol. Any MCP-compatible client can connect by spawning the server process.
Usage
Word must be running with a document open. Tools operate on the active document by default; pass doc="filename.docx" to target a specific open document.
Paragraph and table indices are 1-indexed (matching what you see in Word).
Example workflows
- "Read me paragraph 5" →
get_paragraph(paragraph_index=5) - "Find all mentions of 'confidential'" →
find_text(find="confidential") - "Replace 'old term' with 'new term'" →
replace_text(find="old term", replace="new term") - "Highlight paragraph 20 in yellow" →
set_highlight(paragraph_index=20, color="yellow") - "Show me what page 3 looks like" →
export_page_image(page=3) - "Set paragraph 10 font to bold red" →
set_font(paragraph_index=10, bold=True, color_rgb="FF0000")
Architecture
- COM session per call: Each tool call re-initialises COM (
pythoncom.CoInitialize) and dispatchesWord.Application, which transparently reconnects to the running instance. appparameter injection: Theword_tooldecorator injects the COM app handle as the first argument, but strips it from the public signature so FastMCP doesn't expose it.- Graceful errors: All exceptions are caught and returned as
{"error": "...", "traceback": "..."}JSON dicts — never crash the server. - 1-indexed: Paragraph and table indices match what the user sees in Word.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。