colab-drive-mcp
An MCP server for safely inspecting, editing, and synchronizing Google Colab-compatible .ipynb notebooks through Google Drive.
README
Colab Drive MCP
<!-- mcp-name: io.github.YummyTastyCode/colab-drive-mcp -->
An MCP server for safely inspecting, editing, and synchronizing
Google Colab-compatible .ipynb notebooks through Google Drive.
[!IMPORTANT] This is a notebook file synchronization server, not a notebook execution service. It transfers and edits
.ipynbfiles through Google Drive. It does not connect to Colab runtimes, execute cells, click Run all, keep sessions alive, automate the Colab browser UI, or bypass Colab limits.
Use this MCP when an AI agent needs to prepare a notebook locally, synchronize it with Drive, inspect completed outputs, or return an edited notebook to the user. Open and execute the synchronized notebook separately in Colab, Jupyter, VS Code, or another notebook runtime.
Features
- Inspect and edit notebook cells without loading the entire notebook.
- Create, search, copy, upload, and download notebooks.
- Report live upload and download progress to MCP clients that request it.
- Compare local and Drive notebook copies using Drive metadata and content MD5.
- Keep local file access inside a configured root directory.
- Normalize Colab-specific stream output metadata during downloads.
- Diagnose Google Drive setup without unexpectedly opening a browser.
- Return actionable errors for missing credentials, expired authorization, permissions, missing files, and rate limits.
Non-goals
- Starting, controlling, or monitoring a Google Colab runtime.
- Detecting whether a Colab runtime is connected, busy, or executing a file.
- Executing notebook cells locally or remotely.
- Browser automation, automatic
Run all, or unattended Colab sessions. - Circumventing Colab quotas, idle timeouts, access controls, or usage policies.
- Deploying or running the project contained inside a notebook.
AI Contract
The MCP initialization response includes server-wide instructions that define this project as file synchronization only. Tool descriptions repeat the same boundary where it matters.
manifest.0 provides the same purpose, capabilities, and
non-goals as a typed Zero language contract for agents and repository tooling.
Tools
Setup
get_google_drive_status: diagnose dependencies, credentials, token, and scope.authorize_google_drive: explicitly open the Google OAuth browser flow.
Local notebooks
list_local_notebooks,get_local_notebook,create_local_notebookadd_local_cell,update_local_cell,delete_local_cellsearch_local_cells,clear_local_outputs
Google Drive
list_drive_notebooks,pull_drive_notebook,push_local_notebookget_notebook_sync_status,copy_drive_notebook,get_colab_url
pull_drive_notebook and push_local_notebook synchronize notebook files.
They send MCP progress notifications when the client supplies a progress token,
but clients decide whether and how to display those notifications. They never
execute notebook code. get_colab_url returns a URL but does not open a browser.
get_notebook_sync_status compares the local notebook's upload representation
with Drive's MD5 checksum and returns in_sync, differs, remote_only, or
unknown. It also returns Drive metadata such as modification time and version.
It cannot determine whether Colab is currently executing the notebook because
Google Drive does not expose Colab runtime activity.
Install
Run directly from PyPI with uvx:
COLAB_MCP_ROOT="$HOME/notebooks" uvx --from 'colab-drive-mcp[drive]' colab-drive-mcp
Or install from source:
git clone https://github.com/YummyTastyCode/colab-drive-mcp.git
cd colab-drive-mcp
python3 -m venv .venv
.venv/bin/pip install -e '.[drive]'
Run the server:
COLAB_MCP_ROOT="$HOME/notebooks" .venv/bin/colab-drive-mcp
Local tools can access only .ipynb files below COLAB_MCP_ROOT.
The server is published in the official MCP Registry as:
io.github.YummyTastyCode/colab-drive-mcp
Google Drive Setup
First call get_google_drive_status. It does not open a browser or modify
files. Its response explains the next required step.
To enable Drive:
- Create a Google Cloud project.
- Enable the Google Drive API.
- Configure the OAuth consent screen and add your account as a test user.
- Create an OAuth Client ID with application type Desktop app.
- Download the JSON to
~/.config/colab-mcp/credentials.json. - Call
authorize_google_driveand complete the Google sign-in flow.
The resulting token is stored at ~/.config/colab-mcp/token.json.
Override these locations with:
COLAB_MCP_GOOGLE_CREDENTIALSCOLAB_MCP_GOOGLE_TOKEN
Never commit OAuth credentials or tokens.
Drive Access Modes
The default mode is file, using Google's narrower drive.file scope:
COLAB_MCP_DRIVE_ACCESS=file
This mode can access files created or explicitly opened by this OAuth app. It cannot reliably list all existing notebooks in a user's Drive.
To find and update existing Drive notebooks, explicitly enable full access:
COLAB_MCP_DRIVE_ACCESS=full
Changing access modes may require deleting token.json and calling
authorize_google_drive again.
Codex Configuration
codex mcp add colab-drive \
--env COLAB_MCP_ROOT="$HOME/notebooks" \
--env COLAB_MCP_DRIVE_ACCESS=file \
-- /absolute/path/to/colab-drive-mcp/.venv/bin/colab-drive-mcp
VS Code Configuration
Add this server to the VS Code MCP configuration:
{
"servers": {
"colab-drive": {
"type": "stdio",
"command": "/absolute/path/to/colab-drive-mcp/.venv/bin/colab-drive-mcp",
"env": {
"COLAB_MCP_ROOT": "/absolute/path/to/notebooks",
"COLAB_MCP_DRIVE_ACCESS": "file"
}
}
}
}
Colab Compatibility
Google Colab may add a metadata property to stream outputs. That property is
invalid under the standard nbformat v4 schema. During Drive downloads, this
server removes only that incompatible property. Stream text and all other
outputs are preserved.
Security
- OAuth tokens stay on the local machine.
- Local tools are restricted to
COLAB_MCP_ROOT. - Authorization requires an explicit
authorize_google_drivecall. - Use
fileaccess unless full Drive discovery is required. - Review actions before overwriting an existing Drive file.
This project is not affiliated with or endorsed by Google.
Development
.venv/bin/pip install -e '.[drive,test]'
.venv/bin/pytest
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。