Security

Built so an agent can look, not touch.

Giving an AI agent your Search Console data should not mean giving it the keys to your site. Here is how GSC Magic is built.

Last updated: 2026-10-18

In short

GSC Magic is read-only by default, uses Google OAuth instead of API keys or service-account files, encrypts tokens at rest with AES-256-GCM, checks plan, Google scopes and client group on every tool call, and lets you export or delete everything yourself. Security reports go to security@gscmagic.com.

Read-only by default

Google sign-in asks for Search Console, GA4 and Tag Manager access in one consent. Write tools stay disabled until the workspace owner enables write mode on a paid plan. Member accounts are always read-only, viewers get no tools at all, and access tokens can be minted as read-only with an expiry. The server checks the mode on every request; a client cannot talk its way around it.

OAuth, no secrets to leak

You never create a Google Cloud project, download a JSON key or paste a client secret. Claude, ChatGPT, Claude Code and Cursor connect through OAuth 2.1 with dynamic client registration and PKCE; the consent page is ours and shows exactly which client is asking and where it will return. Refresh tokens rotate with a short reuse window, so a lost response does not log anyone out.

Permissions computed per request

Every tool call is checked against your plan, the Google scopes you actually granted, the keys you added and, on Agency, the client group of the person or token making the call. A member assigned to one client cannot query another client's property, even indirectly: every argument that selects a site is validated against the grant, with exact host matching, so a grant for www.acme.com does not open sc-domain:acme.com.

Data handling

  • Tokens and client keys are encrypted at rest with AES-256-GCM under a keyring that supports rotation.
  • Stored Search Console, Bing and GA4 statistics are tied to your workspace and deleted with it. Export everything as JSON and CSV from Settings at any time.
  • Tool-call logs keep 30 days of metadata (tool name, client, outcome) and no query results.
  • Hosted mode never writes exports to the server's disk.
  • Fetching your robots.txt and sitemaps for the site audit is guarded against private networks and redirect tricks.

Operations

  • HTTPS everywhere with HSTS; strict security headers on the dashboard and the API; no cookies on the MCP endpoint.
  • Rate limits before authentication on the MCP endpoint and webhooks, sensitive-route limits on the API, and sign-in lockouts.
  • Optional two-factor authentication (TOTP authenticator app, with one-time recovery codes) for e-mail and password sign-in. Google sign-in uses your Google account's own second factor.
  • Security events (denied sign-ins, blocked cross-tenant attempts, limiter hits, bad webhook signatures) are logged as one structured event stream.
  • Encrypted, verified database backups with off-site copies; a written incident response runbook with a 72-hour notification target.
  • Dependency audits, secret scanning and an SBOM run on every change.

Google API policy

GSC Magic's use of Google user data follows the Google API Services User Data Policy, including the Limited Use requirements. We use the data only to provide the features you see, we do not sell it, and we do not use it to train models. Details in the privacy policy.

Report a vulnerability

Write to security@gscmagic.com. We acknowledge within 2 business days and share an assessment within 5. Machine-readable details are in /.well-known/security.txt.

Questions

Can the AI agent change my site?

Not by default. Sign-in grants read access; every tool that writes (sitemaps, IndexNow, Bing URL submission, Indexing API) is disabled until the workspace owner turns on write mode on a paid plan and approves the extra Google permission. The server enforces this regardless of what the client asks.

Do you store my Google password?

No. Sign-in happens on Google's page. We store OAuth tokens, encrypted with AES-256-GCM under a rotating key, and you can revoke them from your Google account at any time.

Is my data used to train models?

No. We do not train models and we do not share your data with the AI provider beyond the results of the specific queries your agent runs. Those results are governed by that provider's terms.

Where is the data hosted?

In the European Union. Our processors are listed in the privacy policy.

How do I report a vulnerability?

E-mail security@gscmagic.com. We acknowledge within 2 business days and share an assessment within 5. Good-faith research is welcome; the full policy and scope are in /.well-known/security.txt.