Google Drive¶

Gives agents read-only access to your Google Drive files. Agents can search, list, and read documents, spreadsheets, and other Drive content.

Prerequisites¶

On the host machine (pynchy-server):

# System Chrome — required for GCP Console automation during setup
apt install google-chrome-stable

# Virtual display — required on headless servers for interactive OAuth consent
apt install xvfb x11vnc novnc

Add CHROME_PATH to .env:

CHROME_PATH=/usr/bin/google-chrome-stable

1. Define chrome profile and instance¶

Add a chrome profile and gdrive instance to config.toml:

chrome_profiles = ["mycompany"]

[mcp_servers.gdrive.mycompany]
chrome_profile = "mycompany"

The plugin provides the base spec (Docker image, port, transport, Dockerfile). You only declare the instance with its chrome profile attachment.

2. Grant workspace access¶

[workspaces.mycompany-1]
mcp_servers = ["gdrive.mycompany"]

3. First-time setup¶

Ask your agent to set up Google for the profile:

@Pynchy set up Google for the mycompany profile

The agent calls setup_google(chrome_profile="mycompany"), which automates the full GCP setup flow: create a project, enable the Drive API, configure OAuth consent, create credentials, and run the OAuth authorization. You click "Allow" on the Google consent screen to grant read-only Drive access.

On a headless server, the agent returns a noVNC URL — open it in your browser to interact with the GCP Console and Google login.

4. Verify¶

Trigger a message in a workspace with gdrive.mycompany access. The Docker container starts on-demand:

ssh pynchy-server 'docker ps --filter name=pynchy-mcp-gdrive'

Multiple accounts¶

Each chrome profile maps to one Google account. To access Drive from multiple accounts:

chrome_profiles = ["mycompany", "work"]

[mcp_servers.gdrive.mycompany]
chrome_profile = "mycompany"

[mcp_servers.gdrive.work]
chrome_profile = "work"

[workspaces.mycompany-1]
mcp_servers = ["gdrive.mycompany", "gdrive.work"]

The agent sees separate tool namespaces: mcp__gdrive_mycompany__search and mcp__gdrive_work__search.

Troubleshooting¶

403 errors from Drive API¶

The Drive API isn't enabled for your GCP project. Ask the agent to set up Google for the profile — it will detect the missing API and enable it.

Token expired / authentication errors¶

Ask the agent to set up Google for the profile again. The setup_google tool is idempotent — it detects that credentials exist but tokens are expired, and runs only the OAuth flow.

noVNC not loading¶

Ensure xvfb, x11vnc, and novnc are installed on the host. The setup tools start the virtual display automatically, but the packages must be present.

Browser lock files after crash¶

rm -f data/playwright-profiles/google/SingletonLock
rm -f data/playwright-profiles/google/SingletonSocket
rm -f data/playwright-profiles/google/SingletonCookie

Migration from old gdrive setup¶

If you previously used the old [mcp_servers.gdrive] config with Docker named volumes:

Create the chrome profile directory and move credentials:

mkdir -p data/chrome-profiles/mycompany
cp data/gcp-oauth.keys.json data/chrome-profiles/mycompany/gcp-oauth.keys.json

Update config.toml:
Add chrome_profiles = ["mycompany"]
Remove the old [mcp_servers.gdrive] section (plugin provides it now)
Add instance: [mcp_servers.gdrive.mycompany] with chrome_profile = "mycompany"
Update workspace: mcp_servers = ["gdrive.mycompany"]
Re-authorize (tokens in the old Docker volume won't carry over):
```
@Pynchy set up Google for the mycompany profile
```

Clean up old artifacts:

docker volume rm mcp-gdrive 2>/dev/null
rm data/gcp-oauth.keys.json