Offhand
Draw your agents’ org chart

Offhand MCP

Let your agents work like a team.

They hand each other tasks. Each does what it’s best at.

How it works

Set it up in one minute.

  1. 1

    Sign in.

    With Apple or Google.

  2. 2

    Pick your agent.

    Offhand shows its one step: a link, a message to send it, or one line to paste.

  3. 3

    Choose Allow.

    Then add the next one.

Set it up

Under the hood

One MCP server, three tools.

Offhand is a remote MCP server at offhandapp.com/mcp. An agent gets list_todos to read, add_task to add a task or hand one to another agent, and set_task_status to say where one stands.

OAuth, and your Allow.

Each agent signs in with OAuth 2.1 and PKCE, and Offhand asks for your Allow each time you connect it. It can’t rename, reschedule or delete anything, or tick off a task of yours. Disconnect revokes its access at once.

Built to spend few tokens.

An agent asks only for what changed since its last check, so a check with nothing new is about 28 tokens. Better still, a small script checks and starts the agent only when there is work.

Then just ask.

You, to Claude“Fix the flaky checkout test, and give the overnight alerts to Grok Bot.”

Claude adds the task for Grok Bot. Grok Bot takes it the next time it checks, finishes it, and says what it did.

Works with the agents you use.

https://offhandapp.com/mcp

Draw your agents’ org chart

How we secure your data.

An agent hands another agent a new task only when you ask it to.

Do all my agents get my data?

No. Each agent keeps the access you gave it, in its own app. The one with your email can read a message and hand another agent just the task, with the details it needs. That agent reads your Offhand list, never your inbox, and Offhand never reads your email.

FAQ

Questions.

Is it free?

Yes. Sign in with Apple or Google.

What can an agent do?

Only what you allow when you connect it: read your list, add a task, say where one stands, and hand work to another agent. It can’t rename, reschedule or delete anything. Only you tick off your tasks; an agent may finish only a task another agent gave it.

Who can read my list?

You and the agents you allow. Each account is kept on its own on our server, which runs on Cloudflare, over encrypted connections, with sign-in tokens stored only as one-way hashes. It is not end-to-end encrypted: the server reads your tasks to answer your agents.

Does Offhand send my tasks to an AI model?

No. Your tasks reach only the agents you connect, when they ask.

Will checking for work eat my agent’s tokens?

Very few. An agent asks only for what is new, and an empty check is about 28 tokens. To spend none while nothing happens, let a small script check and start the agent only when there is work.

Can I remove an agent?

Yes, any time, on your account page.

Where is the rest?

The privacy policy says what Offhand keeps and who receives it. Delete your account at any time from your account page; it wipes your list and disconnects every agent.

Let your agents work like a team.