Todos for Agents is a hosted to-do list for AI agents. Agents connect over the Model Context Protocol (MCP), and people use a web interface. Both work with the same tasks. If you're new, start with how it works.
Get access
Self-serve sign-up isn't open yet. When it opens, this section will link straight to it, and you'll get a workspace with your own lists and the connection details for your agent.
To ask about access in the meantime, contact the NimbleBrain team.
Connect an agent
Any AI assistant or agent framework that can connect to a remote MCP server can use Todos for Agents. The steps are the same everywhere:
- Open your workspace and copy its MCP connection details: the endpoint URL and the sign-in method.
- In your assistant or agent framework, add a new MCP server (sometimes called a connector or tool server) and paste the endpoint.
- Complete the sign-in step your client asks for, so the agent acts on your workspace.
- Check the connection by asking your agent: “What's on my to-do list?”
Clients that read a JSON configuration file usually take an entry like this:
Example only. The URL is a placeholder; use the one from your workspace. Field names vary between clients.
{
"mcpServers": {
"todos": {
"type": "http",
"url": "https://<your-endpoint-from-the-dashboard>/mcp"
}
}
}Setup guides for specific assistants will be added here as each one is tested.
Instructions for your agent
Agents discover the task operations on their own, but a few lines of guidance make them use the list consistently. Add something like this to your agent's system prompt or project instructions:
You have access to a persistent to-do list (Todos for Agents).
- At the start of a session, check the list for open and in-progress tasks.
- When you take on multi-step work, write the plan as tasks before starting.
- Mark a task in progress when you start it and done when you finish it.
- If you're waiting on a person, mark the task blocked and say why.
- When asked what's left, answer from the list, not from memory.What agents can do
Once connected, your agent can perform these operations. Each one is described to the agent in plain language when it connects.
| Operation | What it does |
|---|---|
| Create a task | Add one task to a list, with optional priority, due date, assignee, labels, and notes. |
| Plan several tasks | Lay down a multi-step plan in one call. |
| Update a task | Change a task's fields (title, notes, priority, due date, assignee, labels) without changing its status. |
| Change status | Start, complete, block (with a reason), reopen, or cancel a task. |
| List and search | Filter tasks by list, status, assignee, or label, or search titles and notes. |
| Get what's next | Return the highest-priority task that's open and not blocked. |
| Summarize a list | Count tasks per status and flag what needs attention. |
| Manage lists | Create lists and rename or update them. |
Task fields
- Title
- Short description of the work. Required.
- Notes
- Longer details, context, or links.
- Priority
- Urgent, high, normal (the default), or low.
- Due date
- When the task should be done.
- Assignee
- Who owns it: a person's name, or the agent.
- Labels
- Free-form tags for grouping and filtering.
- List
- The list the task belongs to. Tasks go to your default list unless the agent names another.
- Status
- Where the task stands. See below.
Statuses
Every task has exactly one status. Agents change status explicitly, so “done” always means someone marked it done.
- Open
- Not started yet. The default for a new task.
- In progress
- Someone, or the agent, is working on it.
- Blocked
- Waiting on something. Records a reason.
- Done
- Finished. Records when it was completed.
- Cancelled
- Won't be done. Kept for the record.
Tasks can also be archived to hide them from everyday views without changing their status.
Working with people
People use the web interface to see and change the same tasks the agent works on. Anything a person does, whether adding a task, changing a priority, or completing something, is visible to the agent the next time it reads the list. Each task keeps a history of recent activity, so you can see what changed.
A good pattern: let the agent create and update tasks, and have a person review the list each day to reorder priorities and unblock anything waiting on them.
Data and account
Your tasks are stored in the hosted service, which NimbleBrain operates. They're scoped to your workspace; agents connected to a different workspace don't see them.
The privacy policy and terms of service aren't published on this site yet. This section will link to them once they are.
Getting help
Questions about setup or access: contact the NimbleBrain team. Common questions are answered in the FAQ.