Model Context ProtocolCreate a free account
Email MCP server setup for Claude, Cursor and VS Code
BlinkInbox runs a hosted MCP email server. Connect it once and your AI agent can create inboxes, wait for verification emails, and read OTP codes or magic links on its own. New to the idea? Start with email for AI agents.
Remote endpoint (Streamable HTTP)
https://api.blinkinbox.club/api/mcpClients that support OAuth (Claude, Claude Code, Cursor, VS Code) only need the URL: they open a “Sign in with BlinkInbox” window and you approve. Or send an API key as Authorization: Bearer bb_live_…. Same account and limits as the REST API.
1. Add the server to your client
Settings → Connectors → Add custom connector
Name: BlinkInbox
URL: https://api.blinkinbox.club/api/mcp
Click Connect, sign in to BlinkInbox and approve.2. Tools your agent gets
| create_inbox | Create a disposable address (prefix, domain, TTL, label). |
| list_inboxes | Recent inboxes, optionally only active ones. |
| list_messages | Messages in an inbox with detected OTP / verify link. |
| get_message | Full body (text or HTML), codes and links. |
| wait_for_email | Block until a matching email arrives (≤55s per call). |
| get_verification_code | Newest OTP or magic link, waiting if needed. |
| delete_inbox | Delete an inbox and all its mail. |
3. Try a prompt
“Sign up on staging.acme.dev with a fresh BlinkInbox address, confirm the email, and report any step that failed.”
“Run through our password-reset flow at localhost:3000 and tell me if the reset email arrives within 30 seconds.”
“Create an inbox labelled "pricing-alerts", subscribe it to the newsletter on acme.dev, and summarise the first email.”
Email content is written by third parties, so it can contain prompt-injection attempts. Tool results label message bodies as untrusted data. Keep human approval on for sensitive actions. When connecting with OAuth you can grant read-only access (no creating or deleting inboxes) and disconnect any app in Dashboard → Connected apps.