Buzzmark MCP · for AI assistants
Give your AI assistant an inbox.
Buzzmark’s MCP server lets Claude, Cursor and other AI assistants read the email you receive, reply in the same thread and send from your own domain - with approvals, allowlists and limits you set. EU-hosted, and included on every plan.
Works with claude.ai · Claude Desktop · Claude Code · Cursor · any MCP client
Waiting for your approval
Re: When will my order arrive?
To maria@cliente.es · from help@yourshop.com
Just ask
Your email, in plain English.
No dashboards to learn. Tell your assistant what you need and it uses Buzzmark to do it.
“Summarise this morning’s support emails.”
“Draft replies to today’s questions - I’ll approve them.”
“Tell Maria her order ships on Thursday.”
“Who emailed us about refunds this week?”
“Send the weekly numbers to the team.”
“Did my reply to Tom get delivered?”
Set up in a minute
Connect, choose, ask.
-
Connect
In claude.ai or Claude Desktop, add a custom connector with our URL and sign in. For Claude Code or Cursor, create a key on the hive’s AI access page and paste the snippet we give you.
https://buzzmarkapp.com/api/mcp/mail -
Choose what it can do
Pick the hive, the From address on your verified domain, whether it can read incoming mail, and how it may send.
Ask first · Replies only · Listed recipients -
Ask away
Your assistant reads, drafts and sends through Buzzmark. Approve emails in the dashboard, and revoke access whenever you like.
Approve & send ✓
You decide
As much freedom as you’re comfortable with.
Every connection has its own rules. Change them or revoke access at any time - it takes effect on the very next request.
Ask first
The default. Every email waits for a person to approve it, and you get a heads-up email when something is waiting.
Replies only
It can reply to emails the hive received, and nothing else. Ideal for a support inbox.
Listed recipients
Sends straight away, but only to the addresses or whole domains you list.
Anyone
Sends straight away to any address, within your daily limit.
Guard-rails built in
Powerful for your assistant. Boring for attackers.
An assistant that can send email needs firm limits. These come with every connection - there’s nothing to configure.
- One hive, nothing else. No access to other hives, settings or account data.
- A fixed From address. On your verified domain. The assistant can’t change it.
- No BCC, no attachments. And at most 10 recipients per email - it can’t quietly copy mail out or send in bulk.
- Daily and per-minute caps. Your daily limit plus 10 emails a minute, so a runaway agent stops itself.
- Untrusted mail stays untrusted. Incoming email is labelled for the assistant and stripped of hidden characters.
- No reply loops. It won’t answer auto-replies, bounces or no-reply addresses.
- Your rules still apply. Suppressions, verified domains, rate limits and plan allowances, on every send.
- A full audit trail. Every email it sends is tagged and traceable to the connection, and approvals record who approved them.
Questions
Which AI assistants work with it?
Anything that speaks MCP (the Model Context Protocol): claude.ai and Claude Desktop through a custom connector, Claude Code, Cursor, and other MCP-compatible clients and agent frameworks.
Can it send email without me?
Only if you let it. Every connection starts on “Ask first”: emails wait on the hive’s AI access page until a person approves them, and you get an email when something is waiting. You can switch a connection to replies only, a list of allowed recipients, or sending to anyone.
What stops a malicious email from hijacking the assistant?
Several layers. Incoming mail is marked as untrusted for the assistant and stripped of hidden characters; it can’t reply to auto-replies, bounces or no-reply addresses; it only ever sends from the address you chose; and approvals, allowlists and daily limits cap what it can do. A cleverly written email can still mislead an AI, so keep “Ask first” or “Replies only” on for anything sensitive.
Does it cost extra?
No. MCP is included on every plan, the free tier too. Emails an assistant sends count like any other email you send.
Does Buzzmark train AI on my email?
No. Buzzmark doesn’t train anything on your mail. Email the assistant reads is passed to the AI provider you connect, under that provider’s terms - so connect only assistants you trust with it.
Can an assistant manage my account too?
That’s a separate connection. Buzzmark’s account MCP server can create hives and answer questions about logs, bounces and usage, using your account API token. It can’t read or send mail, and mail connections can’t manage the account.
Ready to make some noise?
Spin up a hive, grab a token, and send your first tracked email in about five minutes. The bee’s already warming up the wings.