google-tasks-mcp
Manage Google Tasks from your AI assistant using natural language.
Documentation
Google Tasks MCP Server
Connect an MCP-compatible client to Google Tasks through a private server you run yourself. The server exposes compact tools for reading, searching, summarizing, creating, completing, updating, deleting, and moving Google Tasks.
This project is for self-hosted use. You provide your own Google Cloud OAuth credentials, connect your own Google account, and keep tokens in your own SQLite database.
What You Get
- 19 MCP tools for Google Tasks.
- Local stdio mode for desktop/client-launched setups.
- Streamable HTTP mode for local HTTP or VPS hosting.
- Bearer-token HTTP auth, plus optional OAuth 2.0 gateway mode for MCP clients that support OAuth.
- Compact responses designed for low-context assistant workflows.
- Optional operator-managed multi-account bearer-token routing, for familiar setups such as one personal account and one work account.
Choose A Transport
| Use case | Transport | Auth |
|---|---|---|
| MCP client starts the process directly | `stdio` | No `MCP_BEARER_TOKEN` needed |
| Local HTTP server | Streamable HTTP at `http://127.0.0.1:8787/mcp` | Bearer token |
| VPS or other host | Streamable HTTP at `https://your-domain.example/mcp` | Bearer token or OAuth gateway |
For deeper hosting and distribution guidance, see MCP_SERVER_GUIDE.md and DISTRIBUTION.md.
Install
git clone https://github.com/ebmurha/google-tasks-mcp.git
cd google-tasks-mcp
python3.11 -m venv .venv
. .venv/bin/activate
pip install -e .
cp .env.example .envGenerate a bearer token if you will run HTTP mode:
python -c "import secrets; print(secrets.token_urlsafe(48))"Put the generated value in `.env` as `MCP_BEARER_TOKEN`. Do not commit `.env`.
Google Cloud Setup
1. Create or open a Google Cloud project.
2. Enable the Google Tasks API.
3. Configure the OAuth consent screen.
4. Create an OAuth 2.0 Client ID.
Recommended for local HTTP, VPS, Docker, and other server-style installs:
- Application type: Web application
- Local redirect URI: `http://127.0.0.1:8787/callback`
- Hosted redirect URI: `https://your-domain.example/callback`
- `.env`: set `GOOGLE_CLIENT_ID`, `GOOGLE_CLIENT_SECRET`, and `GOOGLE_REDIRECT_URI`
Local-only alternative:
- Application type: Desktop app
- Download the OAuth client JSON outside this repo.
- Set `GOOGLE_OAUTH_KEYS_PATH` to that file path.
- Leave `GOOGLE_CLIENT_ID` and `GOOGLE_CLIENT_SECRET` empty unless you want env vars to override the JSON file.
Example `.env` for a local web OAuth client:
GOOGLE_CLIENT_ID=your-client-id.apps.googleusercontent.com
GOOGLE_CLIENT_SECRET=your-client-secret
GOOGLE_REDIRECT_URI=http://127.0.0.1:8787/callback
MCP_BEARER_TOKEN=your-generated-token
DB_PATH=./google-tasks.db
BIND_HOST=127.0.0.1
BIND_PORT=8787If the Google OAuth app is in Testing mode, add every Google account you bootstrap as a test user, such as both personal and work accounts.
`GOOGLE_CLIENT_ID` / `GOOGLE_CLIENT_SECRET` / `GOOGLE_OAUTH_KEYS_PATH` identify the Google Cloud OAuth app, not the Google Tasks user account. One OAuth client JSON can be reused for several Google users. Each bootstrap run stores a separate refresh token for the Google account you authorize in the browser.
Bootstrap Google OAuth
Run this once per Google account you want the server to access:
google-tasks-mcp-bootstrapOpen the printed URL, approve access, and paste the authorization code back into the terminal.
For multiple trusted accounts on one HTTP server, create one stored bearer token per account and bootstrap each account separately:
google-tasks-mcp-create-bearer-token --account-id personal --label "Personal account"
google-tasks-mcp-bootstrap --account-id personal
google-tasks-mcp-create-bearer-token --account-id work --label "Work account"
google-tasks-mcp-bootstrap --account-id workUse each printed bearer token only in the matching account's MCP client. The server stores only bearer-token hashes.
Start The Server
HTTP mode:
python -m google_tasks_mcp --transport httpHealth check:
curl http://127.0.0.1:8787/healthzExpected response:
{"ok": true}Stdio mode:
python -m google_tasks_mcp --transport stdioConfiguration check:
python -m google_tasks_mcp --checkConnect An MCP Client
Remote or local HTTP:
URL: http://127.0.0.1:8787/mcp
Auth: BearerFor a VPS, replace the URL with your HTTPS endpoint:
URL: https://your-domain.example/mcp
Auth: BearerLocal stdio:
{
"command": "/path/to/google-tasks-mcp/.venv/bin/python",
"args": ["-m", "google_tasks_mcp", "--transport", "stdio"]
}`MCP_BEARER_TOKEN` is not required for stdio because the MCP client launches the process locally.
Authentication Modes
Bearer-token mode is the default HTTP mode. `/mcp` requires `Authorization: Bearer `.
- `MCP_BEARER_TOKEN` routes to account `default`.
- Tokens created with `google-tasks-mcp-create-bearer-token` can route different clients to different `account_id` values.
- Bearer tokens are displayed once and stored only as hashes.
OAuth 2.0 gateway mode is optional. Enable it when your HTTP MCP client supports OAuth authorization metadata and token refresh.
- Set `MCP_OAUTH_ISSUER`, `MCP_OAUTH_CLIENT_ID`, `MCP_OAUTH_CLIENT_SECRET`, and `MCP_OAUTH_SIGNING_SECRET`.
- Set `MCP_OAUTH_REDIRECT_URIS` to the callback URI values accepted by your MCP client.
- `/mcp` accepts OAuth-issued access tokens and the legacy bearer token.
- OAuth gateway refresh tokens are stored by hash and rotate on use, so clients can reconnect after server restart.
Leave `MCP_OAUTH_REDIRECT_URIS` empty to keep OAuth gateway mode disabled.
Tools
The same 19 tools are available over stdio, bearer-token HTTP, and OAuth gateway HTTP modes. Tools expose standard MCP titles, descriptions, and safety hints where the client supports them.
| Group | Tools | Notes |
|---|---|---|
| Tasklists | `list_tasklists`, `create_tasklist`, `get_tasklist`, `update_tasklist`, `delete_tasklist` | Tasklist delete requires `confirm: true`; non-empty lists require `force: true`. |
| Task reads | `list_tasks`, `get_task` | Read from one tasklist. If `tasklist` is omitted, uses `DEFAULT_TASKLIST` or Google's first list. |
| Task summaries | `today`, `overdue`, `upcoming`, `search`, `digest` | If `tasklist` is omitted, reads all tasklists and includes tasklist context. |
| Task mutations | `clear_completed`, `add`, `complete`, `update`, `uncomplete`, `delete`, `move` | Mutate one tasklist/task at a time. `clear_completed` requires `confirm: true`. |
All `tasklist` arguments accept a tasklist ID or exact title. Task title lookup is exact after trimming whitespace and ignores case.
For `today`, `overdue`, `upcoming`, `search`, and `digest`, omitting `tasklist` reads every tasklist. Returned task objects include `tasklist_id` and `tasklist_title`; `digest` labels items with tasklist context.
For `list_tasks`, `clear_completed`, single-task tools, and write tools, omitting `tasklist` uses `DEFAULT_TASKLIST`, or the first list returned by Google. This prevents unqualified writes from touching every list.
Limitations
These are Google Tasks REST API limits:
- Due dates are date-only. Google drops time-of-day values on task due dates.
- Recurring tasks cannot be created or read through the Google Tasks REST API.
- `clear_completed` hides completed tasks; it does not permanently delete them.
Docker And VPS
Docker:
docker compose up --buildKeep `.env`, OAuth JSON files, and SQLite databases outside images and public bundles.
VPS/systemd/Caddy templates are in `deploy/`:
Replace every placeholder domain, path, and user before deploying.
Troubleshooting
Missing bearer token:
- HTTP `/mcp` requires `Authorization: Bearer ` unless OAuth gateway mode is handling the client.
- Stdio mode does not use `MCP_BEARER_TOKEN`.
Google OAuth app is in Testing mode:
- Add every bootstrapped Google user as a test user.
- Testing-mode refresh tokens can expire after 7 days.
Callback URI mismatch:
- `GOOGLE_REDIRECT_URI` must exactly match an Authorized redirect URI in Google Cloud.
- For local web OAuth, use `http://127.0.0.1:8787/callback` consistently.
Expired or revoked Google refresh token:
- Run `google-tasks-mcp-bootstrap` again for the affected account.
- For multi-account mode, include the same `--account-id` you used before.
OAuth MCP client keeps re-authorizing:
- Ensure the server is running a version with persisted MCP OAuth refresh tokens.
- Check that `DB_PATH` points to persistent storage and survives restarts.
- Verify `MCP_OAUTH_ISSUER` is the public HTTPS base URL with no trailing slash.
More Docs
- MCP_SERVER_GUIDE.md explains hosting models, credential boundaries, and public project vs public service choices.
- DISTRIBUTION.md explains registry, bundle, and directory publishing.
- .env.example lists every supported environment variable.
- google-tasks-mcp-specifications.md is the behavioral source of truth.
Tests
pytestFrequently asked questions
What is google-tasks-mcp?
google-tasks-mcp is Manage Google Tasks from your AI assistant using natural language.
How do I install google-tasks-mcp?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is google-tasks-mcp open source?
Yes — it is hosted on GitHub at https://github.com/ebmurha/google-tasks-mcp and has 2 stars.
Related MCP tools
An AI Gateway, registry, and proxy that sits in front of any MCP, A2A, or REST/gRPC APIs, exposing a unified endpoint with centralized discovery, guardrails and management. Optimizes Agent & Tool calling, and supports plugins.
AI-powered OSINT agent with interactive REPL, MCP server, and CLI. 19 tools. Works with Claude, GPT-4, or local models. For authorized security research only.
Production-grade MCP server giving Claude 27 security intelligence tools across 21 APIs — CVE lookup, EPSS scoring, CISA KEV, MITRE ATT&CK, Shodan, VirusTotal, and more.
Open-source coding agent memory. Records issues, attempts, fixes and decisions, then warns your agent before it repeats an approach that already failed. Native MCP server for Claude Code, Cursor, Antigravity and Codex. 100% local, no cloud, no telemetry. MIT.
🚀 The fast, Pythonic way to build MCP servers and clients Trusted by 19900+ developers. Trusted by 19900+ developers. Trusted by 19900+ developers.
Build effective agents using Model Context Protocol and simple workflow patterns Python-based implementation. Trusted by 7600+ developers.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP