FGAC.ai

Fine-grain access control for AI agents on Gmail, Google Sheets and Google Docs.

smithery badge

Product: https://fgac.ai · Docs: https://fgac.ai/docs · Privacy: https://fgac.ai/privacy

FGAC.ai is a hosted MCP server and API proxy that sits between your AI agents and your Google accounts. Connect one or many Gmail accounts — work, school, personal, and inboxes teammates delegate to you — plus the specific Google Sheets and Docs you choose, and every request passes through deny-by-default access rules you control before it touches Google. Nothing to install, no Google Cloud project needed: sign in with Google and the agent can read that account's mail immediately; sending, editing, and other inboxes are granted from your dashboard or from a one-click approval link the agent hands you when it is denied.

Add FGAC to your agent

MCP endpoint: https://fgac.ai/api/mcp (Streamable HTTP; OAuth 2.1 with dynamic client registration and PKCE; discovery at /.well-known/oauth-protected-resource/mcp).

What you get

• Multiple Gmail accounts — connect several accounts, and let teammates delegate their inboxes to your agent from their own dashboard. Every delegation keeps its own rules and is revocable in one click. No password sharing.
• Editable Google Sheets and Docs — expose individual files read-only or read & write; agents can update cells, append rows, and edit documents there and nowhere else. The rest of your Drive does not exist to them.
• Guardrails — read rules hide sensitive mail (2FA codes, password resets, financial alerts) by label or content pattern; send whitelists limit outbound mail to recipients you approve; permanent deletion is never possible.
• One-click approvals — when an agent needs more (a new recipient, a new sheet), it asks; you approve exactly that grant from a single-use link.
• Nineteen tools — typed Gmail, Sheets, Docs, and comments tools with safety annotations, plus a rule-checked raw Google API escape hatch covering the full Gmail, Sheets, and Docs API surface.
• A request log — every call your agent makes, with what was allowed and what was blocked and why.

Your data is never stored or used for training. See the Terms for service terms.

Licensing

Two different things, two different terms:

• The hosted service at https://fgac.ai is open to anyone — individuals, teams, companies, and schools — under its Terms of Service and Privacy Policy. Connecting an agent through fgac.ai does not involve this repository's license at all.
• This source code is licensed for personal use only: an independent individual may self-host it to manage their own Gmail or Google Workspace accounts. Use by or on behalf of a company, government agency, or educational institution, and offering the code as a hosted or multi-tenant service, are prohibited without a separate license from the author. See LICENSE for the exact terms and liability limitations.

Developing

This is a Next.js app deployed on Vercel, with Clerk for authentication and Neon Postgres via Drizzle. There is exactly one supported way to run it locally, and it depends on access to the project's Vercel environment:

npx vercel link --yes --project fine-grain-access-control   # once per clone
npx vercel env pull .env.local --environment=development     # dev Clerk + Neon creds
npm run db:branch                                            # isolated Neon branch
npm run dev

Node 20.9+ is required. npm run env:check diagnoses environment problems. Contributor rules, QA workflow, and database safety guards are documented in CLAUDE.md and docs/.