# Connect Claude to Orbital

> Add Orbital AI to Claude or Claude Code, sign in, approve the connection for one workspace and project, and start asking Claude to build and backtest strategies.

Source: https://orbitalaip.com/docs/claude-and-chatgpt/connecting

> [!IMPORTANT] Disclaimer
> Backtest results are hypothetical and do not predict future performance. Nothing here is investment advice: Orbital computes the rules you supplied and recommends nothing. Orbital is not a registered investment adviser or broker-dealer. Costs are as modelled by the chosen cost profile and may differ from live trading.

## Before you start

You need an Orbital account, which you can [create](https://orbitalaip.com/docs/getting-started/sign-up) at [app.orbitalaip.com/signup](https://app.orbitalaip.com/signup). You also need the **Backtesting** app turned on in the project you will use, since every tool call checks it before it runs.

A new account starts with Backtesting on in its first project, but a workspace you create later starts without it; see [Switch an app on or off in a project](https://orbitalaip.com/docs/workspace-and-account/apps#switch-an-app-on-or-off-in-a-project).

## Two ways in

Orbital AI for ChatGPT is coming, and there is nothing to install yet; see [ChatGPT](https://orbitalaip.com/docs/claude-and-chatgpt/chatgpt).

### Claude on the web or desktop

1. **Add the connector**
   
   Open [Add to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=Orbital%20AI&connectorUrl=https%3A%2F%2Fmcp.orbitalaip.com%2Fmcp). It opens Claude's **Add custom connector** form for **Orbital AI** at `https://mcp.orbitalaip.com/mcp`. Or go to **Customize** → **Connectors**, click **+ Add**, then **Add custom connector**, and enter the name and the URL. Click **Continue**, review the detected authentication settings, and click **Continue** again.

2. **Set up sign-in**
   
   Under **Authentication**, choose **Sign in now**. Under **OAuth client**, keep **Use Claude's published identity (Recommended)**. Orbital needs no request headers. Click **Add**.

3. **Approve in Orbital**
   
   Claude sends your browser to Orbital's consent page, described below.

4. **Switch it on in a chat**
   
   Click **+**, then **Connectors**, and switch **Orbital AI** on for that conversation.

> [!NOTE] Managed Claude accounts
> If an organization manages your Claude account, an owner adds the connector under **Organization settings** → **Connectors**: **Add**, then **Custom**, then **Web**, with the same name, URL and sign-in settings, ending with **Add**. You then open **Customize** → **Connectors** and click **Connect** next to **Orbital AI**.

### Claude Code

Add the server from your terminal.

```bash title="Terminal"
claude mcp add --transport http orbital-ai https://mcp.orbitalaip.com/mcp
```

Then run `/mcp` in Claude Code and follow the browser steps to sign in and approve the connection.

## Signing in and approving

Both ways end on Orbital's consent page, titled **Connect to Orbital**. If you are signed out, Orbital shows **Sign in** first, with **Sign up** for a new account, and brings you back afterwards.

A pending request stays valid for 30 minutes; after that the page says **Nothing to connect** and you [start again](https://orbitalaip.com/docs/claude-and-chatgpt/troubleshooting#the-consent-request-expired).

The consent page shows:

- a sentence naming who is asking and where you go back to, such as "Claude (claude.ai) wants to connect to Orbital and will be sent back to claude.ai"; from Claude Code it names localhost and the port Claude Code is listening on;

- the permissions asked for (the table below);

- a **Workspace** picker and a **Project** picker, which start on your current workspace and project;

- the binding sentence: "The connection is bound to the project _name_. To use another project, connect again.";

- two buttons: **Approve** and **Deny**.

| Permission | What the page says |
| --- | --- |
| read | Read your strategies and backtests |
| write | Create and edit strategies |
| run | Run backtests, which spends credits |
| openid, email | See your email address |
| offline_access | Stay connected |

There is no per-permission choice: **Approve** grants the list as shown and returns you to Claude, and **Deny** returns you with nothing connected.

Write and run also need a project role that may write; see [Permissions and refusals](https://orbitalaip.com/docs/claude-and-chatgpt/tools#permissions-and-refusals). With a read-only role, Claude can still list and read, and for anything else it relays the [role refusal](https://orbitalaip.com/docs/claude-and-chatgpt/troubleshooting#your-role-in-this-project-cannot-change-strategies-or-spend-credits): "Your role in this project (_role_) cannot change strategies or spend credits."

## One workspace, one project

The connection is bound to the workspace and project you chose. Claude's tools reach only that project: its strategies, versions and runs, and nothing elsewhere in your workspace.

To work in another project, connect again and pick it on the consent page:

- In Claude, [remove the connection](https://orbitalaip.com/docs/claude-and-chatgpt/connecting#removing-the-connection), then add **Orbital AI** again.

- In Claude Code, run `/mcp`, choose **orbital-ai**, then **Clear authentication**; run `/mcp` again and sign in.

## What Claude can and cannot do

Claude can:

- list the strategies in the project and read each one with its latest backtest verdict;

- create strategies from rule cards or as a Program, save changes as new versions, and build a Program's code;

- estimate a backtest, run it, read the results and the trade log, and cancel a run that is still going.

Claude cannot:

- place orders, deploy a strategy or reach Trade; trading stays inside Orbital;

- delete a strategy, a version or a run, because no tool exists for it;

- reach another project or workspace, or do more than the permissions you approved;

- recommend anything: results are hypothetical and Orbital computes only the rules you supplied.

[The eleven tools](https://orbitalaip.com/docs/claude-and-chatgpt/tools) lists every tool.

## Three prompts to start with

- "Build a 50/200-day golden-cross strategy on SPY and backtest it over the last three years."
- "Backtest an RSI mean-reversion rule on AAPL, MSFT and NVDA with a 5% stop since 2022."
- "Show my strategies and the verdict of the latest backtest of each."

## Removing the connection

In Claude:

1. **Open Customize → Connectors**
   
   

2. **Find Orbital AI**
   
   It is listed under **Connectors**.

3. **Remove it**
   
   Click **Remove**, or open the three-dot menu next to it, and follow the prompts.

In Claude Code, sign out of the server first, which revokes its tokens at Orbital, then remove it from your terminal.

```bash title="Terminal"
claude mcp logout orbital-ai
claude mcp remove orbital-ai
```

Orbital's settings do not list connected assistants, so the removal happens in Claude.

## Help

Read [Troubleshooting](https://orbitalaip.com/docs/claude-and-chatgpt/troubleshooting) first, then email [hello@orbitalaip.com](mailto:hello@orbitalaip.com) or ask on [Discord](https://discord.gg/XwmVs84tcd).