onenote-mcp-python
A standalone Model Context Protocol server for Microsoft OneNote desktop, connecting via the Windows COM API through a PowerShell bridge to enable page CRUD, search, export, and organization tools without an API key.
README
OneNote MCP (Python)
A standalone Model Context Protocol server for Microsoft OneNote desktop, written in Python. It connects to OneNote through its Windows COM API via a bundled PowerShell bridge and does not require an API key or an embedded AI provider.
Requirements
- Windows
- Microsoft OneNote desktop from Office (comes with Microsoft 365; not the Store app), with at least one notebook open
- Python 3.11+ and uv — install uv with:
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
(uv manages Python itself, so a separate Python install is not required.)
Quickstart
git clone https://github.com/eddyficial/onenote-mcp-python.git
cd onenote-mcp-python
uv sync
uv run onenote-mcp-setup
onenote-mcp-setup configures Codex, Claude Desktop, and Claude Code in one
shot, preserving unrelated MCP entries. Preview with --dry-run, or target one
client with --client codex, claude-desktop, or claude-code (--client claude covers both Claude clients). Reload the client and the onenote_*
tools appear.
Run manually
uv run onenote-mcp
The server speaks MCP over stdio; all diagnostics go to stderr.
Connect an MCP client by hand
If you'd rather not use the setup command:
Claude Desktop
Add to %APPDATA%\Claude\claude_desktop_config.json:
{
"mcpServers": {
"onenote": {
"command": "uv",
"args": ["--directory", "C:\\path\\to\\onenote-mcp-python", "run", "onenote-mcp"]
}
}
}
Claude Code
claude mcp add onenote -- uv --directory C:\path\to\onenote-mcp-python run onenote-mcp
Codex
Add to %USERPROFILE%\.codex\config.toml:
[mcp_servers.onenote]
command = "uv"
args = ["--directory", "C:\\path\\to\\onenote-mcp-python", "run", "onenote-mcp"]
Capabilities
The server exposes 27 onenote_* tools covering hierarchy, page CRUD, search,
organization, rich content, export, safe deletion, knowledge digests, action
and decision extraction, duplicate and stale-page health checks, weekly review
source packs, and preview-first templates.
| Tool | What it does |
|---|---|
onenote_hierarchy |
List notebooks, section groups, sections, and pages with IDs |
onenote_get_page |
Read a page's title and flattened text |
onenote_search |
Full-text search across pages |
onenote_knowledge_digest |
Source-grounded executive/detailed digest across a scope |
onenote_extract_insights |
Extract action items, owners, due dates, decisions, risks, questions |
onenote_health_report |
Audit duplicates, stale/untitled/empty pages, ownerless actions |
onenote_template_preview |
Preview a trusted page template (read-only) |
onenote_create_from_template |
Create a page from a template (preview-first) |
onenote_weekly_review |
Build a weekly-review source pack (read-only) |
onenote_create_page |
Create a page in a section |
onenote_append_page |
Append text to a page |
onenote_update_page |
Replace or append a page's body |
onenote_insert_rich_content |
Append XHTML fragments and/or images |
onenote_rename_page |
Set a page's title |
onenote_move_page |
Move a page to another section (returns new page ID) |
onenote_reorder_pages |
Change page order within a section |
onenote_create_section |
Create a section |
onenote_create_section_group |
Create a section group |
onenote_create_notebook |
Create a notebook |
onenote_rename_section |
Rename a section |
onenote_move_section |
Move a section to another parent |
onenote_reorder_sections |
Change section tab order |
onenote_navigate |
Open an object in the visible OneNote window |
onenote_export |
Export a page/section to pdf, html, docx, mhtml, xps, or onenote |
onenote_delete_page |
Delete a page (recycle bin by default) |
onenote_delete_section |
Delete a section (recycle bin by default) |
onenote_delete_notebook |
Delete/close a notebook (recycle bin by default) |
Why a PowerShell bridge instead of pywin32?
With x64 Click-to-Run Office, the OneNote COM typelib is registered only under
the Win32 registry key, so 64-bit Python COM dispatch (pywin32/comtypes) fails
with TYPE_E_LIBNOTREGISTERED. .NET's COM binder is unaffected, so the server
spawns bridge\onenote_bridge.ps1 once (powershell.exe -NoProfile -NonInteractive -ExecutionPolicy Bypass -File ...) and speaks a small
JSON-lines RPC over its stdio: {id, op, args} requests, {id, ok, result, error} responses, with a 60-second per-call timeout. The bridge is the single
canonical COM path.
Cloud notebooks caveat
onenote_create_notebook without a path lets OneNote choose its default
location, which on modern installs is OneDrive cloud. A just-created cloud
notebook can reject immediate writes with COM error 0x80042030 until it
syncs. For reliable scripted workflows, pass an absolute local path (e.g.
C:\Users\you\Documents\Notebooks) — local notebooks accept section and page
writes instantly.
Safety
- Template creation previews by default.
- Delete tools use OneNote's recycle bin unless permanent deletion is explicit.
- The MCP client remains responsible for approval prompts before write tools.
Test
uv run pytest
Other implementations
Same 27 tools, same schemas — pick your runtime:
- onenote-mcp-dotnet — C#/.NET, direct COM (no bridge), ships a self-contained exe (easiest install)
- onenote-mcp-windows — Bun/TypeScript original, PowerShell bridge, no build step
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。