How to Use Claude With SelfManager: Connect, Set Permissions, and Manage Tasks

You can connect Claude to SelfManager through its MCP server, then ask Claude to read your real work and make permitted changes. You sign in to SelfManager, approve the connection, and choose its permissions.

You need an active SelfManager plan; the seven-day trial counts. The SelfManager MCP page has the current setup steps.

Start with one useful question about today. Once that works, try a small task update before asking for a larger plan.

What connecting Claude actually does

MCP stands for Model Context Protocol. It lets an assistant work with a connected service through tools.

In this setup, Claude can request information from SelfManager rather than asking you to paste your task list into the conversation. Changes depend on the permissions you approve.

Reading access can include tables shared with you. A read-only connection prevents changes; it does not limit reading to one table you name in a prompt.

This is also different from selecting a Claude model inside SelfManager. The MCP connection lets you work from Claude’s interface. SelfManager’s model menu controls the AI features you use inside SelfManager itself.

Before you connect

Have these ready:

• An active SelfManager plan or trial.
• Access to custom connectors in your Claude account.
• One recognizable table to use for your first check.

Anthropic currently documents custom remote connectors for Free, Pro, Max, Team, and Enterprise. Free accounts are limited to one custom connector. Organizational accounts may require an authorized owner or administrator to add the connector first.

Check the account you are signed into on both sides. This matters if you use separate personal and work accounts.

Step 1: Add SelfManager as a custom connector

In Claude on the web or desktop:

1. Open the Connectors area. Depending on the interface, look under Settings or Customize.
2. Choose Add custom connector.
3. Name it SelfManager.
4. Enter this remote server address:

https://api.selfmanager.ai/mcp

1. Add the connector and select Connect.
2. Sign in on the SelfManager approval page.
3. Review the permissions, then select Allow.

Anthropic’s current instructions describe adding a custom connector by name and server URL. SelfManager supplies the endpoint above and its sign-in flow.

Use the server address exactly as written. It is different from the public SelfManager website address.

Step 2: Choose what Claude may do

For a first connection, read-only is a reasonable way to learn what Claude can retrieve. If you want it to add or complete tasks, enable changes.

The current approval options separate basic changes from access to additional content and actions.

Reading is always enabled. Make changes is selected by default; the other permissions listed above require a separate choice.

A prompt such as “only use my Client A table” helps specify the task. It does not replace the connection’s access controls.

Step 3: Check the connection with a reading request

Try an original prompt like:

Read today’s SelfManager tables. List their names, unfinished tasks, and recorded deadlines. Do not change anything.

Compare the response with SelfManager.

Check whether Claude found the right day and whether the task names and progress match. An empty day may simply mean your work is recorded elsewhere.

Then make the request more specific:

Open the Client A table dated October 19, 2026. Explain what remains unfinished and identify any unanswered questions in its notes.

Using a date and table name reduces ambiguity. Replace the example with your own date and table.

SelfManager keeps work on individual days, so make the intended date explicit when similar table names appear repeatedly.

Step 4: Try one small change

If you enabled changes, start with something you can immediately inspect.

For example:

In my private Client A table dated October 19, 2026, add one task: “Confirm the final homepage headline.” Leave the existing tasks unchanged.

After Claude reports success, open that table and check the result.

For a completion update:

Mark “Confirm the final homepage headline” completed in that table.

These are examples of the table creation, task addition, and completion workflows SelfManager supports through connected assistants.

If several tasks match, resolve the ambiguity before continuing. A short clarification is easier than correcting the wrong task.

Practical prompts for daily work

The most useful prompts name the source, desired result, and whether anything should be saved.

Decide what to work on

Read today’s tasks and deadlines. Suggest a realistic order based on the information recorded. Explain any assumptions and leave everything unchanged.

Use this when your task list is clear enough to compare but you want help choosing an order.

Turn a brief into proposed actions

Read the available text of the brief attached to my Client A table. Draft the next actions and list anything that needs clarification. Show the draft before adding tasks.

Document access requires its permission. Through MCP, Claude receives extracted document text, not the original file.

Prepare tomorrow’s work

Review the unfinished tasks in today’s Client A table. Propose what belongs tomorrow, what needs clarification, and what can wait. Ask before creating tomorrow’s table.

This makes carrying work forward a decision instead of assuming every unfinished task should become tomorrow’s obligation.

Review a recorded period

Review October 12 through October 16, 2026. Separate completed work, partial progress, and unresolved questions. Use only recorded evidence and identify gaps.

Claude can read work across dates and use it to prepare its own summary. That is different from running SelfManager’s built-in AI Review.

Save a review you have checked

Save the review we just agreed on to my SelfManager AI History.

SelfManager supports saving the assistant’s conversation or review to AI History with its attribution.

If you already use Claude for thinking through client work, begin with one real table. Connect it and try one reading request before expanding the workflow.

What Claude cannot do through this connection

The connection does not expose every SelfManager action.

Current restrictions include sharing tables, changing access roles, rewriting existing notes or comments, and running SelfManager’s own AI Plan or AI Review.

It also cannot hear audio recordings through MCP. Supported transcripts are separate from the audio itself.

Deletion is possible only with the separate permission and within the documented restrictions. It is limited to one item at a time, generally in private tables; a whole-table deletion requires a preview. Do not rely on older descriptions saying deletion is impossible.

Some actions are flagged for Claude to ask first. Your Claude tool-approval settings also affect that behavior.

For your initial setup, choose the permissions needed for the work you intend to do.

How to check what changed

SelfManager marks assistant-written comments with the app’s name. Table changes appear in the table log when logging is enabled, with attribution such as “via Claude.”

The Assistant activity page keeps a 30-day list of connected-app changes. Use it to check where a change happened, then inspect the relevant table for the result.

After a larger request, verify:

• The intended dates and tables.
• Whether anything was duplicated.
• Task completion and recorded time.
• Any shared comments or deadlines.

A fluent reply is useful context. The saved work is the result to check.

What uses SelfManager AI credits?

The connection uses SelfManager’s existing AI credit balance.

Reading a single day or table and updating existing tasks are among the zero-credit operations. Searches, reading across dates, and creating new content can consume credits. Larger additions have batching rules.

A conversational request can involve several tool calls, so “one message” does not necessarily mean “one credit.” Check AI Usage to see charged requests.

Your Claude account’s own limits still apply separately. A SelfManager connection does not replace your Claude account.

Common connection problems

SelfManager asks you to sign in again. Approval requires a recent sign-in. Sign out, sign back in, and retry.

The first connection attempt fails. SelfManager’s guide says an idle sign-in service may need a brief warm-up. Wait about ten seconds and try again.

An action is unavailable. Compare the requested action with the permissions you approved.

To change permissions, remove and reconnect the connector, reviewing the new approval screen.

Disconnecting Claude

In SelfManager, open My Profile → Connected apps and disconnect the connection. SelfManager says access ends within about a minute.

The assistant’s provider handles information it has already received under its own terms. Disconnecting prevents further access; it should not be treated as deleting earlier Claude conversations.

Start with one day and one table

A useful first session is simple:

1. Read today’s work.
2. Check the result.
3. Add or complete one task if changes are enabled.
4. Inspect the saved update.

Once that works, try a client brief, tomorrow’s plan, or a review of recorded work.

SelfManager gives Claude a structured place to read from and write to. Clear requests help turn that access into useful daily work.

Frequently asked questions

Do I need a paid Claude plan?

Anthropic currently supports custom remote connectors on Free and paid plans, with one custom connector allowed on Free. Team and Enterprise access can depend on organizational setup.

Can I connect during the SelfManager trial?

Yes. The seven-day trial counts as an active plan for MCP.

Can I make the connection read-only?

Yes. Turn off Make changes during approval. Reading remains enabled and follows the supported access available to your account.

Can Claude run SelfManager AI Review?

No. Claude can read your recorded work and write its own review, but SelfManager’s built-in AI Review runs inside SelfManager.

Can I see which changes Claude made?

Yes. Check attributed comments, enabled table logs, and Assistant activity. The activity list covers the last 30 days.