Before connecting your inbox to any AI tool, you should be able to answer a short list of questions about how it handles auth, what permissions it requests, whether it stores your email content, and what controls exist over write actions. Most people skip this step. This checklist gives you the specific things to verify — and explains how Connect My Email addresses each one — so you can make an informed decision rather than a hopeful one.
1. Authentication: OAuth vs. App Passwords
What to look for: The connector should support OAuth where possible. OAuth means you authorize access through your email provider's own consent screen — you never hand your password to a third party. For providers that don't support OAuth for third-party connectors (some IMAP-only setups), app-specific passwords are the correct fallback: a generated credential with limited scope that can be revoked independently of your main account password.
Red flags: Any connector that asks for your account password directly. Any connector that stores credentials in plaintext or in your local config files.
CME approach: CME uses OAuth with standard flows for Gmail, Outlook, and other major providers that support it. For providers where OAuth isn't available for third-party apps, CME uses app-specific passwords. Your main account password is never requested or stored.
2. Permission Scope: Least Privilege
What to look for: The connector should request only the permissions it actually needs. A read-only use case — summarizing email, searching for senders, running analytics — has no reason to request write permissions. If a connector asks for full read/write/send/delete access by default, that's a scope overreach.
Questions to ask:
- Does the connector separate read and write permissions?
- Can I grant read-only access and upgrade later?
- What does the OAuth scope string actually say?
CME approach: CME separates read permissions from write permissions by plan tier. Free accounts get read/search/sweep access only — no write scope is requested during OAuth. Write capabilities (draft, reply, send, move, delete, flag) are gated behind Pro and Power plans, and the OAuth grant expands accordingly when you upgrade. You're not pre-authorizing write access before you've decided you want it.
3. Data in Transit
What to look for: All communication between the connector and email providers — and between the connector and the AI model — should be encrypted in transit using current TLS standards. This is table stakes, but verify it explicitly.
CME approach: All data in transit is encrypted. Connections between CME's servers and email providers use TLS. Connections between your AI assistant and CME's MCP endpoint use HTTPS. There's no unencrypted path.
4. Email Content Storage and Retention
What to look for: Does the connector store your email content? For how long? Where? What happens when you disconnect?
Many connectors need to cache data temporarily to execute tool calls. That's expected. What you're evaluating is whether the connector is building a persistent copy of your inbox — a secondary store that lives on after you've disconnected.
Questions to ask:
- Is email content persisted beyond what's needed to execute a tool call?
- What's the retention policy for cached data?
- When I revoke access, what's deleted?
CME approach: Email content is not stored as a persistent secondary index. Tool calls read from your live inbox and return results — there's no long-lived copy of your messages on CME's infrastructure. Revoke the OAuth grant from your provider's connected-apps settings, and access ends.
5. Write Action Gates
What to look for: Write actions — sending email, archiving, deleting, moving — should never happen silently or automatically. A well-designed connector requires explicit user confirmation before any action with side effects is executed.
This is especially important for deletion and sending. Sent emails can't be unsent. Permanent deletes can't be undone. A connector that executes these operations without a review step is a liability.
What good looks like:
- The AI presents a summary of what it intends to do before doing it
- The user confirms before execution
- Bulk destructive actions (delete, archive at scale) offer a simulation step that shows impact without committing
CME approach: CME's write tools are designed around the simulate-then-confirm pattern. Your AI will describe what it intends to do, surface a preview (including simulate_cleanup for bulk operations), and wait for explicit confirmation. Nothing is sent, deleted, or moved without user approval at that moment in the conversation.
6. Audit Logs and Visibility
What to look for: You should be able to see what actions have been taken through the connector. This is important for both personal accountability (did that cleanup run do what I expected?) and for team or shared-inbox use cases where multiple people may be involved.
CME approach: CME surfaces what actions were taken within the conversation context. For review purposes, you can ask your AI assistant to summarize what operations were performed in a session.
7. Revocation
What to look for: Revoking access should be simple and immediate. You should be able to disconnect the AI from your inbox without contacting support or waiting for a delayed process. The standard mechanism for OAuth connectors is revoking the grant from your email provider's connected-apps settings — this should be sufficient.
CME approach: Revoke the OAuth grant from your provider's connected-apps page (Google Account settings, Microsoft account settings, etc.) and CME's access is immediately terminated. No additional step required.
The Quick Checklist
Before connecting any AI email connector:
- Authenticates via OAuth or app-specific password — not your main account password
- Requests minimal permissions; separates read and write scope
- All data encrypted in transit
- No persistent secondary copy of your email content
- Write actions require explicit confirmation before execution
- Bulk destructive actions have a simulate/preview step
- Revocation is immediate via provider settings
- You can see what actions were taken
If a connector can't answer yes to all of these, that's useful information.
FAQ
Is OAuth always more secure than an app-specific password?
OAuth is generally preferred because the grant is scoped and revocable through your provider's standard interface, without changing your main password. App-specific passwords are a reasonable fallback for providers that don't support OAuth for third-party apps — they can be generated with a specific label and revoked independently. Both are better than sharing your actual account password.
What does "least privilege" mean in practice for email connectors?
It means the connector should request only the permissions required for the features you're using. A connector that needs to read email doesn't need the ability to send or delete. Least privilege limits the potential impact if something goes wrong — the connector can only do what it was explicitly authorized to do.
How do I check what OAuth permissions I've granted?
For Gmail: visit myaccount.google.com → Security → Third-party apps with account access. For Microsoft/Outlook: visit account.microsoft.com → Privacy → Apps and services. You'll see what scope was granted to each connected app and can revoke at any time.
What's the difference between archiving and deleting in the context of CME?
Archiving moves email out of the inbox but retains it in your account, searchable and recoverable. Deleting removes the message (though most providers have a trash/deleted folder with a recovery window). CME defaults to archive-first language — when your AI assistant "cleans up" email, the default action is archive unless permanent deletion is explicitly requested and confirmed.
Should I use a separate email account for testing AI connectors?
That's a reasonable approach if you're cautious. Using a low-stakes secondary account for initial testing lets you evaluate the connector's behavior without risking anything in your primary inbox. Once you've confirmed it behaves as expected, you can connect your main account.
What if the connector I'm evaluating doesn't publish answers to these questions?
Absence of documentation is itself a signal. Any connector designed with security in mind should be able to clearly answer how it handles auth, what it stores, and how revocation works. If you can't find that information, reach out and ask directly — or treat the gap as a reason for caution.
Connect My Email is built on these principles. See exactly what's available on the tools page, review pricing to understand what permissions each plan requests, or start with the free email cleaner — read-only access, no write permissions required.