Inboxmcp

What inboxmcp does

The inboxmcp MCP server makes mail from connected inboxes available to an AI assistant. It supports incoming-message access, saved preferences, processing state, decision recording, and optional outgoing email. Reading incoming mail is separate from sending: the agent can prepare a plain-text draft, but the signed-in owner must review the recipients and content and confirm delivery in InboxMCP.

The service does not classify messages or notify users on its own. The connected AI decides which messages matter and whether the user should be alerted. Incoming access is read-only with respect to the source mailbox, so the service does not send, delete, move, or mark source messages as read.

How it works

Connect an existing mailbox through a supported provider preset or a publicly reachable IMAP server using certificate-verified TLS on port 993. An app password or provider-specific client password may be required. Collection begins from a new-mail baseline after the mailbox is saved; older messages are not imported automatically.

An AI client connects to https://app.inboxmcp.ai/mcp using Streamable HTTP and OAuth with PKCE. The email:read permission is required. Separate permissions control processing progress, decision updates, and sending. Access is mailbox-scoped, and newly added mailboxes do not automatically become available to existing grants.

Each connection can maintain its own checkpoint and batch acknowledgements in InboxMCP. Completed events are tracked, while unfinished events remain pending. A failure between analysis and acknowledgement may result in an event being processed again. Batch results include saved preferences and a view link for each mail record, and incomplete or unavailable messages are identified rather than represented as complete.

Setup and configuration

Create or sign in to an InboxMCP account, add a mailbox, and verify the connection. Enable automatic collection when saving the mailbox, or leave it disabled to start later. Then connect the inboxmcp MCP server from a compatible AI client, sign in, and authorize only the intended mailboxes and permissions.

For ongoing monitoring, grant consumer-state:write in addition to email:read. Sending requires the separately approved email:send permission; it is unticked by default. Use the dashboard's setup request or the AI-readable setup guide to configure the client's supported event subscription or scheduled task. Installing the connection alone does not create an automatic check.

Test with a new email sent after the collection baseline is established. Have the AI process it, inspect the original-message link, and run another check to confirm that the completed event is not repeated.

Tools and capabilities

The inboxmcp MCP server supports capabilities for:

• Reading authorized incoming email.
• Recording independent processing progress and acknowledgements.
• Listing consumer receipts with list_consumer_receipts.
• Applying saved preferences to returned mail context.
• Recording agent decisions when the permission is granted.
• Preparing and sending owner-confirmed plain-text email when sending is authorized.
• Using the set_up_email_monitoring and check_new_email MCP prompts where available.

Limitations and notes

The repository contains documentation and registry metadata rather than the hosted server source code. The hosted service is described as a limited public beta. Native Gmail and Microsoft authorization is only available when enabled by the deployment; the current setup can use IMAP instead.

The service encrypts credentials, webhook keys, message bodies, drafts, and saved preferences at rest, but it is not end-to-end encrypted because authorized work requires the service to decrypt data. Mail retention defaults to 30 days. Incoming email does not directly wake Claude; a supported scheduled task is required for that workflow. Optional Grok delivery is separate from independent collection, and collection does not require a Grok account, webhook, or external state service.

Permissions must be deliberately reapproved to add access. Existing grants and refresh tokens do not widen automatically. Deferrals do not block new mail, but deferred messages require review in the dashboard.