Connect a new tab dashboard to ChatGPT: a practical guide

To connect a new tab dashboard to ChatGPT, you need a dashboard that exposes supported tools, a compatible ChatGPT account and an authorized connection. A bookmark that opens ChatGPT does not do this. The connection has to let ChatGPT request information or perform specific actions on your dashboard.

Dashy V2 is developing that connection through MCP. This is a preparation guide, not a public installation walkthrough. V2 is planned for October 2026; the original Dashy store release does not include the complete new AI package. We will publish the public connection instructions when they are ready.

Decide what you want ChatGPT to do

There are three different needs hiding inside “connect my dashboard.”

A shortcut is the simplest choice for the first case. You can add a link to ChatGPT wherever your start page supports links, and keep using your existing account there.

The third case is where integration earns its place. It can remove the manual step between deciding what to do and updating the tool where the work lives. It also needs more care, because the assistant is changing saved information.

What MCP contributes

The Model Context Protocol defines a way for AI applications to connect to servers that expose tools and context. In a dashboard integration, those tools might list widgets or update an eligible task.

MCP is not a promise that every assistant can operate every part of an app. The server supplies particular capabilities, and the client has to support them. Authentication and granted access still matter.

For Dashy, the public AI guide is the place to check supported preview workflows. A tool connection to Dashy is also different from choosing an OpenAI model inside Dashy Agent: one starts in ChatGPT, while the other starts in Dashy.

Check ChatGPT access before troubleshooting the dashboard

OpenAI's current MCP documentation distinguishes read access from write actions and describes plan and workspace requirements. At this review, full write support is documented for Business and Enterprise/Edu on the web, while Pro access is described more narrowly. Availability can change; check that page and your workspace settings before buying a plan for this purpose.

A missing create-app control may therefore be an account or administrator restriction. It does not automatically mean the dashboard endpoint is broken.

What the setup should establish

For a released integration, follow the dashboard provider's official connection instructions. In ChatGPT's documented custom-app flow, an authorized user supplies the server details, completes authentication where required and scans the tools. Use the provider's published endpoint, not an address copied from an old development log. OpenAI's setup reference explains the current controls.

Before using your real workspace, establish three things:

1. Identity: the consent page belongs to the product you intended to connect, and you are signing into the right account.
2. Access: the connection grants the actions you need, without assuming unrelated files or dashboards are included.
3. Destination: you know which dashboard and item the assistant will operate on.

Do not paste passwords, API keys or session tokens into the chat. Authentication belongs in the documented authorization flow.

Start with a read, then one small change

Once a supported connection is available, a useful first prompt is:

List the dashboards this connection can access. Do not change anything.

Then narrow the request:

On my Study dashboard, list the notes and task widgets. Tell me which actions you can perform on them through this connection.

These are suggested prompts, not a guarantee that every connection exposes those operations. If a capability is missing, the assistant should explain that instead of pretending to have used it.

For the first write, use a clearly named disposable item:

If this connection supports creating a note, add one note titled “Connection check” to the Study dashboard containing “Read the chapter outline.” Leave all existing notes unchanged.

Check the dashboard. Open the note. Confirm its content and location. Reopen the page once to establish that the result persisted. This is a much clearer trial than asking ChatGPT to “make everything better.”

Write requests that identify the target

“Move the weather” becomes ambiguous when there are two weather widgets. Prefer a description that includes the dashboard, the current item and the destination:

On my Home dashboard, move the Tokyo weather widget below the clock. Keep the London weather widget where it is. If that space is occupied, show me a proposed arrangement first.

The exact action must still be supported. For consequential changes, the product may require approval or block the operation. Do not turn an unavailable confirmation into a request to bypass permission checks.

If the result is missing, avoid creating a second problem

Read the current state before repeating a write. A request can finish on the server while the visible page is behind, or the assistant can receive an error and describe it badly. Retrying an add operation without checking may create a duplicate.

Keep a short record: the requested action, target dashboard, approximate time and exact error. Exclude credentials and private note contents. That gives support something useful to investigate.

For everyday use, connect the smallest amount of context that makes the task possible. Our AI dashboard privacy guide covers that choice, and the Claude workflow guide applies the same discipline to a different assistant.