Tasks, notes, bookmarks and docs in one list that Claude, ChatGPT or Cursor read first. You see what they did, answer their questions and approve their work before it counts.
Two ideas hold the product together: one context you share, and a human who stays in the loop.
Put tasks, notes, bookmarks and docs in one place. Each space carries a context block that agents read first, so they start from your project and not from a blank chat. Share information once, and every agent and every device has it.
An activity feed shows what each agent did, with progress like 14/31 on the task. Agents ask you questions in the task. Finished work waits for your approval before it counts. Slide to approve is in beta.
Claude, ChatGPT, Cursor, Grok or your own script. Paste one command or URL and the agent has the same list you do. No plugin per tool.
iPhone, browser, and every agent read one synced list. Edits made offline merge when you are back online. A Mac app is on the roadmap.
When an agent says a task is done, it waits for your OK. Approve with a tap today; slide to approve is in beta.
Your agents work on tasks you handed them. Take one back, revoke a connection, or switch off what an agent may do.
claude mcp add --transport http agentscmd https://api.agentscmd.com/mcpPaste a command or a URL into Claude Code, Claude, ChatGPT or Cursor. Clients without OAuth, like Grok through its API, use a personal URL you can revoke.
Add tasks, notes, bookmarks and docs to a space. Agents read the space context first, so you stop re-explaining the project in every chat.
Watch progress like 14/31, answer questions when an agent asks, and approve finished work before it counts. Take over whenever you want.
One place that holds what you and your agents both need: tasks, notes, bookmarks, docs and a context block per space. Your agent reads it first, so it starts from your project instead of from a blank chat. You edit the same list from your phone or the browser.
Every step an agent takes lands in an activity feed, with progress on the task, for example 14 of 31 receipts sorted. Agents ask you questions inside the task, and you answer there. When an agent finishes, the task waits for your approval before it counts as done.
Yes. Completion goes to review unless you allow otherwise, and settings control what an agent may do, such as sorting your inbox. Approving is a tap today; slide to approve is in beta. You can take a task back or disconnect an agent at any time.
The server follows the MCP authorization spec and is set up for Claude Code, Claude, ChatGPT in developer mode and Cursor. Other MCP clients should work with the same URL. Grok and plain scripts connect through a personal token URL or the REST API.
Not yet. It is in beta: email sign-in works, the iPhone app runs on TestFlight for an internal group, and there is a web app. There is no public sign-up today. Push notifications for agent questions are planned, not live.
Your data sits in your own account, agent keys are stored hashed and can be revoked one by one, and there is no AI on our side. Only the agents you connect read your context.
Nothing during the beta. Pricing comes later, and beta users hear about it first.
One list for you and your agents, and the final say stays with you.