Bring your agents

Your agents are welcome. On your terms.

Connect Claude, or an agent you built, through Rookery's MCP server or REST API. Each key joins as a named agent with only the permissions you choose, and everything it does shows up under its name.

Get early access See how it works

Read-only API and MCP from Pro, $24 a month. Agents that act come with Business. 14-day free trial, no card needed.

Activity

Bookings bot drafted a reply to KateYour agent · waiting for Jess's yes

Quill drafted a reply to PriyaFollowed Refunds v7 · waiting for your yes

Bookings bot tagged it BookingYour agent

Bookings bot may

ReadWork on conversationsDraft replies

“I already use Claude for half my work. I just don't want to hand it the keys to my whole mailbox.”

Daniel, a consultant who works on his own

Before, and with Rookery

Let them in. Keep the keys.

Access

BeforeGive an agent your mailbox password and hope.

With RookeryA key with only the permissions it needs: read, work on conversations, draft or send.

Who did what

BeforeEverything looks like it came from you.

With RookeryThe key shows up as its own agent, by name, in conversations and the audit log.

Sending

BeforeThe agent sends whatever it decides.

With RookeryWithout the Send replies permission it can only draft, and a person approves.

Copy and paste

BeforeYou paste emails into a chat window and back.

With RookeryIt reads, adds notes, tags, assigns, drafts and asks for documents directly.

How it works

Three steps.

Make a key.In Settings, an owner or admin names it and picks what it may do.
Connect.Point Claude Code or another MCP client at Rookery's MCP server, or call the REST API.
Watch it work.It shows up by name, and its drafts wait in Needs your yes, like Quill's.

What an agent can do

Through MCP or the REST API at /api/v1, with a key that starts rk_:

Read: list_conversations, get_conversation, list_mailboxes, list_tags, search_documents.
Work on conversations: add_note, set_status, tag_conversation, assign_conversation.
Reply: reply drafts for a person's yes, or sends with the Send replies permission.
Ask for documents: request_documents, with Fetch's upload link and reminders.
Each conversation comes with Scout's risk flag and reasons, so your agent can tell a likely scam.

What it doesn't do yet

The honest bit

Sending is a permission you give. A key with Send replies can send on its own, after a 10-minute undo window, without a person's yes. Leave it off and every reply is a draft.
Your agent reads email as it was written. Quill treats words inside an email as text, never as orders. Your own agent gets the conversation plus Scout's risk flags, so build it to be careful too.
Keys only, for now. Signing in with OAuth, for Claude and other connectors, is coming. Coming soon
Connections to your other tools, like accounting and CRM, are coming. Coming soon

Works with

Your yesNothing goes out without your yes, and every send has an undo.See how it works → Teach it onceWrite how you like things done, in plain words. The helpers follow it and say so.See how it works → Scout says whyScam warnings that explain themselves in plain words.See how it works →

Questions

Which AI can I connect?

Any MCP client, like Claude Code, or anything that can call a REST API with a key.

Can it send email on its own?

Only with the Send replies permission. Then a reply goes out after a 10-minute undo window. Without it, replies wait for a person.

Can I take access away?

Yes. Revoke the key in Settings and it stops working.

Which plans include it?

Read-only API and MCP come with Pro and Team. Agents that act, by sending, assigning and tagging, are on Business.

When can I start?

We're inviting a small group first. Request early access and we'll reply personally.

Get your evenings back from your inbox.

It takes ten seconds. We're letting people in a few at a time, so we can set each one up properly.

14-day free trial · no card needed · founding members get 50% off the first year