Google APIs for agents and terminals

gum gives agents and terminal users a controlled path into Google APIs. It is one Go binary, one local credential model, and one dispatch path for CLI calls, MCP tool calls, and sandboxed code.

The surface is catalog-first. Instead of minting a top-level command for every Google product, gum lets the caller search the embedded catalog, inspect the operation contract, then dispatch through gum read, gum write, or gum destructive.

Try it

bash

# Find Google API operations by task.
gum search "gmail messages"
gum describe gmail.users.messages.list

# Register a Google OAuth client, then authorize a small scope set.
printf '%s' "$GOOGLE_OAUTH_CLIENT_SECRET" \
  | gum auth use-oauth-client --client-id "$GOOGLE_OAUTH_CLIENT_ID" --secret-stdin
gum login --service gmail,calendar

# Make a read-class API call through the local profile.
gum read gmail.users.messages.list --args '{"userId":"me","maxResults":5}' --output json

# Install agent skills and MCP config after previewing the write plan.
gum setup --dry-run
gum setup --target codex --features skills,mcp --yes

What gum covers

gum v1.0 ships with 222 catalog operations across 32 services. It covers Gmail, Calendar, Drive, Docs, Sheets, Slides, Tasks, Admin, Vault, Chat, Meet, Classroom, Forms, Apps Script, People, Photos, YouTube, Search Console, Google Ads, Maps, Custom Search, and bundled plugin services such as Flights, Scholar, Patents, Trends, and YouTube transcripts.

The operations by service pages are generated from the embedded catalog. They are the product-level entry point. The command index is generated from gum schema --json and documents the CLI surface.

What gum does

Keeps the MCP tool roster small, then loads operation details only after

gum.search_apis or gum search finds a candidate.

Resolves Google credentials from a local profile instead of putting refresh

tokens, service-account keys, API keys, or Ads developer tokens into prompts.

Sends read, write, and destructive calls through separate commands and MCP

tools. The dispatcher rejects a call when the operation risk class and command do not match.

Shapes large Google responses before they hit an agent context window. Field

masks, compact formats, and output profiles are part of the normal workflow.

Installs agent skills and MCP client config for Codex, Claude, Cursor, and

Gemini after showing the files it plans to write.

Pick your path

First install: Install, then Quickstart.
Product lookup: Operations by service and Service coverage.
Auth setup: Auth and the service-specific auth guides.
Agent setup: MCP server, Agent setup, and Automation.
Safe calls: API workflows, Safety, and Output shaping.
CLI flags: generated Command index.
Local state and tests: Paths and profiles, then Live testing.
Plugins: Plugins and the plugin contract.

Project

gum is open source under the MIT license. It is not affiliated with Google. Use it only with Google accounts, projects, data, and APIs you are authorized to operate.