google-tasks-mcp-server
MCP server for Google Tasks: full read/write access to your task lists and tasks (create, update, complete, delete, move, real due dates) from Claude Desktop, Claude Code, Codex, Cursor, or any MCP client. TypeScript, local OAuth, no third-party backend.
Documentation
google-tasks-mcp-server
An MCP server for Google Tasks. Gives any MCP client (Claude Desktop, Claude Code, Codex, Cursor, etc.) full read/write access to your task lists and tasks: list, create, update, complete, delete, and move them, with real due dates.
There's no official Google Tasks MCP server, so this covers that. Full CRUD, OAuth installed-app auth flow (your credentials never leave your machine), TypeScript throughout.
Tools
Task lists
- `google_tasks_list_tasklists` list all task lists
- `google_tasks_get_tasklist` get one task list
- `google_tasks_create_tasklist` create a new task list
- `google_tasks_update_tasklist` rename a task list
- `google_tasks_delete_tasklist` delete a task list and all its tasks
Tasks
- `google_tasks_list_tasks` list tasks in a list, with due-date/updated filters and pagination
- `google_tasks_get_task` get one task
- `google_tasks_create_task` create a task (optionally as a subtask, optionally positioned)
- `google_tasks_update_task` update title/notes/due date/status
- `google_tasks_complete_task` / `google_tasks_reopen_task` mark done / not done
- `google_tasks_delete_task` delete a task
- `google_tasks_move_task` reorder, re-parent, or move a task to a different list
- `google_tasks_clear_completed_tasks` clear all completed tasks from a list
Every tool supports `response_format: "markdown" | "json"`.
Quick start
From npm:
npm install -g @girmmy/google-tasks-mcp-serverOr don't install it at all and let your MCP client run it via `npx` (see step 4).
Or from source:
git clone https://github.com/girmmy/google-tasks-mcp-server.git
cd google-tasks-mcp-server
npm install
npm run buildEither way, follow Setup below to create your own Google OAuth credentials and authorize. There's no shared or hosted version of this, it talks directly to your own Google account.
Setup
1. Create a Google Cloud OAuth client
Each user needs their own OAuth client. It's not something that can be shared, since anyone using your client secret could impersonate your app (though they'd still need each individual user's own consent).
1. Go to the Google Cloud Console, create (or pick) a project.
2. APIs & Services → Library → enable the Google Tasks API.
3. APIs & Services → OAuth consent screen (Google now calls this "Google Auth Platform"):
4. Clients → Create Client → Application type Desktop app. Create it.
5. Grab the client ID and client secret, either via "Download JSON" right after creation or from the client's detail page later. Google's console won't show a secret's plaintext again once you navigate away, only download or regenerate it, so if you lose it just add a new secret instead of hunting for the old one.
6. Save that file. By default this server looks for it at:
`~/.config/google-tasks-mcp-server/client_secret.json`
(override the path with the `GOOGLE_TASKS_CLIENT_SECRET` env var, useful if you'd rather keep it inside the project directory, e.g. `./credentials/client_secret.json`. `credentials/` is already gitignored.)
The file should look like:
{
"installed": {
"client_id": "YOUR_CLIENT_ID.apps.googleusercontent.com",
"client_secret": "YOUR_CLIENT_SECRET",
"auth_uri": "https://accounts.google.com/o/oauth2/auth",
"token_uri": "https://oauth2.googleapis.com/token",
"redirect_uris": ["http://localhost"]
}
}2. Install
From npm (nothing to build):
npm install -g @girmmy/google-tasks-mcp-serverFrom source:
npm install
npm run build3. Authorize (one-time)
If you installed from npm:
google-tasks-mcp-auth
# or without installing anything:
npx -y -p @girmmy/google-tasks-mcp-server google-tasks-mcp-authFrom source:
npm run auth
# or, if your client secret lives at a custom path:
GOOGLE_TASKS_CLIENT_SECRET=./credentials/client_secret.json npm run authThis prints a Google consent URL and tries to open it in your default browser. Sign in, grant access, and it redirects back to a short-lived local server on `http://localhost:53682`. The token gets cached to `~/.config/google-tasks-mcp-server/token.json` (override with `GOOGLE_TASKS_TOKEN_PATH`) and refreshed automatically from then on, so you shouldn't need to run this again unless you revoke access.
Run this on the same machine and in the same regular terminal you'll actually use day to day. See Troubleshooting below if you're running it from a container, VM, or sandboxed shell where `localhost` doesn't map back to your browser.
4. Run it
npm startOr wire it into an MCP client. For Claude Desktop / Claude Code, add to your MCP config (usually `~/Library/Application Support/Claude/claude_desktop_config.json` on macOS).
Using npm, with your client secret at the default `~/.config/...` path, this is the whole config:
{
"mcpServers": {
"google-tasks": {
"command": "npx",
"args": ["-y", "@girmmy/google-tasks-mcp-server"]
}
}
}Or pointing at a source checkout:
{
"mcpServers": {
"google-tasks": {
"command": "node",
"args": ["/absolute/path/to/google-tasks-mcp-server/dist/index.js"],
"env": {
"GOOGLE_TASKS_CLIENT_SECRET": "/absolute/path/to/google-tasks-mcp-server/credentials/client_secret.json"
}
}
}
}Omit the `env` block if you saved your client secret at the default `~/.config/...` path. Fully quit and reopen your MCP client afterward.
Codex
Codex uses TOML rather than JSON. Add this to `~/.codex/config.toml`:
[mcp_servers.google-tasks]
command = "npx"
args = ["-y", "@girmmy/google-tasks-mcp-server"]
startup_timeout_sec = 60Or let the CLI write it for you:
codex mcp add google-tasks -- npx -y @girmmy/google-tasks-mcp-serverRaise `startup_timeout_sec` if the first launch times out. A cold `npx` downloads the
package before the server can answer `initialize`, and Codex's default allowance is short
enough that this sometimes trips on the first run only.
If your client secret isn't at the default path, add:
[mcp_servers.google-tasks.env]
GOOGLE_TASKS_CLIENT_SECRET = "/absolute/path/to/client_secret.json"Check it registered with `codex mcp list`. Authorization is the same one-time
`google-tasks-mcp-auth` step as everywhere else — run it in a normal terminal, not inside
Codex, since it needs to open a browser and catch a loopback redirect.
Environment variables
| Variable | Default | Purpose |
|---|---|---|
| `GOOGLE_TASKS_CLIENT_SECRET` | `~/.config/google-tasks-mcp-server/client_secret.json` | Path to the downloaded OAuth client JSON |
| `GOOGLE_TASKS_TOKEN_PATH` | `~/.config/google-tasks-mcp-server/token.json` | Path where the cached OAuth token is stored |
| `GOOGLE_TASKS_OAUTH_PORT` | `53682` | Loopback port used only during `npm run auth` |
Security notes
- The client secret and cached token are plain files on disk (mode `0600` for the token). Never commit them (`.gitignore` already excludes `client_secret*.json`, `token.json`, `credentials/`).
- The server requests the full `https://www.googleapis.com/auth/tasks` scope (read/write). For read-only access, edit `TASKS_SCOPE` in `src/constants.ts` to `.../auth/tasks.readonly` before running `npm run auth`. Write tools will then fail with a 403 from Google, which is expected.
- Tool annotations mark deletion/clear operations as `destructiveHint: true` so MCP clients can warn or gate on them.
- Nothing here talks to any server but Google's own APIs and, during `npm run auth`, a local loopback listener. No third-party backend involved.
Troubleshooting
`npm run auth` opens a URL, I click Allow, and it just hangs.
The process running `npm run auth` isn't on the same machine/network as the browser you're clicking Allow in. Common if you're running the server inside a container, remote dev environment, or sandboxed shell that isolates `localhost`. Fix: run `npm run auth` directly in a normal local terminal on the machine whose browser you're using.
`Error [TransformError] ... You installed esbuild for another platform`
Your `node_modules` was installed on a different OS/architecture than you're running on now, e.g. copied from a Docker container or a different machine. Fix: `rm -rf node_modules package-lock.json && npm install`.
`Authorization failed: invalid_client`
Your `client_secret.json` has the wrong client secret in it, usually from copy-pasting it by hand instead of downloading it directly. Go back to Google Cloud Console, add a fresh client secret, download or copy it directly, and update your `client_secret.json`.
Server exits immediately with "No cached Google Tasks token found"
You haven't authorized yet, or the flow didn't finish. Run `google-tasks-mcp-auth` (npm install) or `npm run auth` (source checkout) before starting the server.
Development
npm run dev # run directly from TypeScript with auto-reload
npm run build # type-check and compile to dist/Test with the MCP Inspector:
npx @modelcontextprotocol/inspector node dist/index.jsContributing
Issues and PRs welcome. This covers the full Tasks API surface but hasn't been tested across every edge case, so if you hit a bug or want a feature, open an issue.
License
MIT, see LICENSE.
Frequently asked questions
What is google-tasks-mcp-server?
google-tasks-mcp-server is MCP server for Google Tasks: full read/write access to your task lists and tasks (create, update, complete, delete, move, real due dates) from Claude Desktop, Claude Code, Codex, Cursor, or any MCP client. TypeScript, local OAuth, no third-party backend.
How do I install google-tasks-mcp-server?
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-server open source?
Yes — it is hosted on GitHub at https://github.com/girmmy/google-tasks-mcp-server.
Related MCP tools
Open-source cross-agent memory layer for coding agents via MCP. Compatible with Claude Code, Codex, Cursor, Windsurf, Gemini CLI, Antigravity, OpenClaw, Hermes Agent, Oh-my-Pi, Pi, Copilot, Kiro, OpenCode, and Trae.
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 go-to web for your AI coding agent — local-first search, fetch, crawl & research over MCP. No API keys, no cloud, $0/query. Public beta.
AI Skills, MCP Tools, and CLI for Unity Engine. Full AI develop and test loop. Use cli for quick setup. Efficient token usage, advanced tools. Any C# method may be turned into a tool by a single line. Works with Claude Code, Gemini, Copilot, Cursor and any other absolutely for free.
Cut AI token costs 95%+ on code exploration. The leading MCP server for precise, symbol-level GitHub code retrieval via tree-sitter AST. Works with Claude Code, Cursor & any MCP client. 313B+ tokens saved.
Official remote MCP server for Atlassian. Securely connect Jira, Confluence, Jira Service Management, Bitbucket, and Compass to Claude, ChatGPT, Cursor, VS Code, and other AI tools using OAuth 2.1 or API tokens.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP