# How to connect an AI agent to your task list

*Last updated 2026-10-05 · Matthew Blode*

Connect Claude to your Done Bear tasks through MCP. Copy the setup command, sign in, choose your workspace, and check a read and a write.

## The short answer

> Connect an AI agent to your task list by adding Done Bear’s hosted MCP server, signing in, and approving the connection. Claude can then read and update the tasks you can access. Check the workspace before making changes. Done Bear is a task manager with a shared list for people and agents.

## Key takeaways

- Done Bear’s hosted MCP endpoint is https://mcp.donebear.com/mcp.
- Claude Code connects with one terminal command followed by OAuth sign-in.
- Claude’s custom connector setup uses the same hosted endpoint.
- A connected agent works with the tasks your account can access.
- A connection check needs a returned task and a saved change you can inspect.

---

Connecting your task list gives an assistant a way to work with the same tasks
you read in the app. You can ask for a summary of upcoming work, capture a task
from a conversation, or turn a project discussion into next actions. The list
stays the place where you inspect what was saved and choose what to do next.

This guide starts with Claude and a small connection check. Keep the first
request read-only, then create a task whose title you can recognise. That gives
you a concrete result to inspect before asking an assistant to organise work
that matters to you. The same check applies when you change clients later.

## Choose the list you want to connect

Choose the Done Bear account and workspace that contain the tasks you want the
agent to use. Open the app and check the workspace name before starting setup.
If you keep personal tasks and shared work in separate workspaces, decide which
one the first request should use. Write that name into your test prompt.

A connection and a workspace are different choices. The connection establishes
which account the client can use; the workspace chooses the list for an
operation. [Done Bear’s MCP documentation](/docs/mcp/get-started)
explains the hosted connection. Keep the app open beside Claude so you can
compare the returned tasks with the intended list rather than relying on memory.

Use your existing account. If you need a new one, [create a Done Bear account](/register)
before approving the connection. Start with a workspace containing a few tasks
that you can recognise. You do not need to move your whole system or create a
new set of projects to test an assistant’s access to your list.

## Add Done Bear to Claude Code

Add the hosted HTTP server from a terminal using the command in
[Done Bear’s setup instructions](https://donebear.com/docs/mcp/get-started):

```bash
claude mcp add --transport http donebear https://mcp.donebear.com/mcp
```

Then open Claude Code and run `/mcp` to inspect the connection and complete any
requested authentication. [Anthropic’s MCP reference](https://code.claude.com/docs/en/mcp)
covers HTTP setup and connection status. The command registers a server; it does
not by itself prove that your account is signed in or that a task can be read.

Keep configuration scope in mind if you work in several code projects. Start
from the project where you want the assistant to use the list, and consult the
client’s scope options before sharing configuration across projects. You can
repeat the same read check in another project later. Avoid pasting access tokens
into prompts when the browser sign-in flow is available.

## Add the connector in Claude

For Claude’s web or desktop interface, open Settings, then Connectors, add a
custom connector and enter this URL, following
[Anthropic’s remote connector guide](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp):

```text
https://mcp.donebear.com/mcp
```

Use a name you recognise, such as Done Bear. Follow the sign-in and connection
controls shown by Claude. Organisation settings can affect which connectors you
can add, so check your client’s displayed requirements if the custom connector
option is missing. Do not substitute the app’s homepage or API hostname for the
MCP endpoint: they serve different routes.

Choose one connection method for the first test. If you have configured the
same service through several clients, identify which client made the returned
request before diagnosing a failure. A working desktop connection is useful
evidence for that client, but you should still check Claude Code separately if
that is where you intend to use the task list while coding.

## Approve access and read one task

Complete Done Bear’s sign-in and approval flow, then ask the agent for a bounded
read. [The MCP setup documentation](https://donebear.com/docs/mcp/get-started)
describes OAuth authentication; you do not need a local Done Bear server for
this path. Check the account and client shown during approval before continuing.
After approval, return to the client that initiated the connection.

Try this prompt, replacing the workspace name with yours:

```text
List my available Done Bear workspaces. In the workspace named Personal,
show up to five open tasks. Do not change any tasks.
```

Check a returned title against the app. If the workspace is missing, stop the
write test and establish which account approved the client. If a read fails,
record the client’s error and reconnect through that client’s connection
controls. Repeating a broad request such as “organise everything” gives you
less information about where setup failed than a small read does.

## Save a task and inspect it in the app

Create one recognisable task after the read succeeds. The
[task tools reference](https://donebear.com/docs/mcp/tools) lists the available
operations. Give the workspace, title and intended view together so the agent
has enough information to save the task where you expect. Keep the request
small enough that you can check the result without searching several projects.

```text
In my Personal workspace, create a task named “Check my agent connection”
in Inbox. Return its task number and title. Do not change other tasks.
```

Open Inbox in the same workspace and inspect the saved title. Check the returned
task number as well if your client displays it. A response claiming success is
less useful than a saved task you can inspect. If the task appears elsewhere,
ask for its recorded view before making another change, then correct the
specific task using its number.

Once you have inspected the task, decide whether to keep it as a setup reminder
or complete that test task. Avoid using an existing deadline or shared task as
your first write test. [Open your Done Bear list](/login) to inspect the result;
you can use your existing account without setting up another workspace for the
connection check.

## Give the agent a bounded next request

Name the workspace, the tasks and the permitted change when you ask an agent to
help with your list. A request to draft next actions can leave the list
unchanged until you have read the draft. A request to save agreed actions can
name the project and view. Both make the result easier to compare with your
intent than a request to “clean up my tasks”.

For example, ask the agent to read a project and suggest three next actions,
without saving them. Review those suggestions and choose which ones belong on
your list. Then ask for those tasks to be created in the named project. When a
task describes work outside Done Bear, give it a checkable result rather than
using completion of the tool call as the done condition.

## Choose another client when it fits your work

Use the client you already work in, then repeat the same read and write checks.
[Done Bear’s client setup](https://donebear.com/docs/mcp/get-started) covers other
MCP clients. The endpoint remains the same; configuration and authentication
controls vary by client. A connection in one client is not evidence that a
second client is ready, even when both use the same account.

If you are choosing a task manager before choosing an agent, compare the list
and device support first. The [task manager comparison](/compare) can help you
check that fit. You can keep using your existing list without an agent when
manual capture and a weekly review already serve your needs. Connect an
assistant when you have a specific request that benefits from reading or
updating that same list.

## Keep the task list useful after setup

Done Bear is a shared task list for people and connected agents. Its
[MCP tools](https://donebear.com/docs/mcp/tools) let an assistant work with tasks,
projects and comments through the hosted connection. You remain responsible for
choosing what belongs on the list and inspecting whether the requested work
was delivered. Keep a completion condition beside work that needs evidence
from a document, deployment or external service.

A connection is optional. If your current workflow is working, there is no need
to add another client to maintain. If you do connect one, start with the bounded
read, check one saved task, and expand the scope as results warrant it. Return
to the [weekly review guide](/guides/how-to-run-a-weekly-review) when you want to
check the list itself rather than its connection.


## Frequently asked questions

### Can I connect Claude without installing the Done Bear CLI?

Yes. The hosted MCP server is separate from the Done Bear CLI. Add the endpoint to Claude, then complete the browser sign-in and approval. You do not need to run a local Done Bear server or copy an API key into a chat. Claude Code itself must already be installed for the terminal instructions.

### Can an AI agent update tasks as well as read them?

Yes. Done Bear’s MCP tools include task creation, editing, completion and comments, alongside read operations. The account and granted permissions determine access. Start with a read, then a disposable task you can inspect in the app. Give the agent a bounded instruction before letting it change existing work or several tasks at once.

### Why can Claude see a different task list from my app?

Usually, the account or workspace differs. Ask Claude to list available workspaces and name the workspace used for the next operation. Compare that name with the workspace open in Done Bear. Reconnect if the approved account is wrong. A successful connection message alone does not show that the agent has selected the list you intended.

### Does a connected agent finish my work automatically?

No. The connection lets an agent use task tools when instructed; it does not prove the work described by a task is complete. Decide what evidence would satisfy the task before asking the agent to close it. For a document, inspect the document. For a deployed change, check the running result in the intended environment.

### Can I use an agent other than Claude?

Yes. Done Bear exposes a hosted MCP server, and its setup documentation includes Cursor, VS Code, Codex and other clients. Each client has its own connection and authentication controls. Follow the instructions for the client you use, then repeat the workspace, read and write checks here. Availability can also depend on your organisation’s client policy.


## Where this information comes from

- [Done Bear, MCP setup instructions (the hosted endpoint, supported clients and OAuth setup)](https://donebear.com/docs/mcp/get-started)
- [Done Bear, MCP tools (task, workspace, project and comment operations)](https://donebear.com/docs/mcp/tools)
- [Anthropic, Connect Claude Code to tools via MCP (adding an HTTP server and checking authentication in Claude Code)](https://code.claude.com/docs/en/mcp)
- [Anthropic, Custom connectors using remote MCP (adding a remote custom connector in Claude)](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp)

---

Every Done Bear page is listed at https://donebear.com/llms.txt. Any page has a markdown copy at its URL plus `.md`. Start from the map rather than guessing a path.