# Connect your AI

Give your AI assistant or coding tool its own cloud computer in about two minutes. You add one URL, sign in to WBX Sandbox, and it can start sandboxes, run code, jobs and batches, and read the results.

Your MCP server URL `https://sandbox.wbx.cz/mcp`

## No coding needed

- [Claude](#claude): claude.ai, desktop and mobile app
- [ChatGPT](#chatgpt): Plus, Pro, Business, Enterprise
- [Tutorial](https://sandbox.wbx.cz/docs/tutorials.md): step by step with screenshots

## For developers

- [Claude Code](#claude-code): one command
- [Cursor](#cursor): mcp.json
- [VS Code](#vscode): GitHub Copilot
- [Codex CLI](#codex): codex mcp add
- [Gemini CLI](#gemini): gemini mcp add
- [LangChain](#langchain): Deep Agents backend
- [Other clients](#other): HTTP or stdio
- [Python SDK](https://sandbox.wbx.cz/docs/sdk.md): no MCP at all

## Claude

Works on every Claude plan; the Free plan allows one custom connector. On Team and Enterprise plans an organization Owner adds it once (*Organization settings → Connectors → Add → Custom*) and everyone then just clicks *Connect*.

1. ### Open Customize → Connectors and click *Add custom connector*

   In claude.ai or the Claude desktop app.

   ![Claude settings, Customize – Connectors, with the Add menu open showing Add custom connector](https://sandbox.wbx.cz/img/docs/claude-customize.webp)
2. ### Paste the URL and click *Continue*

   Name it **WBX Sandbox**, paste `https://sandbox.wbx.cz/mcp`, click *Continue* and confirm. Leave the advanced OAuth settings empty – Claude registers itself. [Screenshots of every step](https://sandbox.wbx.cz/docs/tutorials.md#no-code)

   ![The Add custom connector dialog with the name WBX Sandbox and the MCP server URL filled in](https://sandbox.wbx.cz/img/docs/claude-dialog.webp)
3. ### Sign in to WBX Sandbox and click *Allow access*

   Claude opens our sign-in page. Approving connects your team (only team owners and admins can approve).
4. ### In a chat, switch it on and ask

   Click **+** → *Connectors* and turn on **WBX Sandbox**. Then ask for something that needs a computer – see the [prompts below](#prompts).

   ![The + menu in a Claude chat with Connectors open and the WBX connector switched on](https://sandbox.wbx.cz/img/docs/claude-composer.webp)

## ChatGPT

Custom MCP connectors need Developer mode, available on Plus, Pro, Business, Enterprise and Edu. In a workspace an admin may need to allow it first. OpenAI renames these menus from time to time – if a label differs, look for the same steps nearby.

1. ### Turn on Developer mode

   *Settings → Apps & Connectors → Advanced settings → Developer mode*.
2. ### Create the connector

   Back in *Apps & Connectors* click *Create*. Name: **WBX Sandbox**, URL: `https://sandbox.wbx.cz/mcp`, Authentication: **OAuth**. Click *Create*.
3. ### Sign in to WBX Sandbox and click *Allow access*

   Same as above – ChatGPT opens our sign-in page.
4. ### Start a new chat and ask

   Make sure WBX Sandbox is enabled for the chat (from the **+** menu), then ask. ChatGPT asks you to confirm before it runs actions.

## Try these prompts

“Use WBX Sandbox to make me a Snake game I can play in my browser, and send me the link.”

“Use WBX Sandbox: which month in this spreadsheet had the best sales? Draw me a chart.”

“Clone https://github.com/psf/requests, run the test suite and tell me what fails.”

“Run these 20 scripts as a batch with at most 5 at a time and a $1 limit, then summarise the results.”

## For developers

Every client below signs in with OAuth in the browser – you approve access on our page and no key is stored in a config file. Where a browser isn't available (a server, CI, SSH), use an API key from *API keys* instead; each section shows how. Approving a connection and creating API keys needs a team owner or admin.

### Claude Code

```
claude mcp add --transport http wbxsandbox https://sandbox.wbx.cz/mcp
```

Then run `/mcp` inside Claude Code and sign in through the browser.

**Without a browser:** add `--header "Authorization: Bearer YOUR_API_KEY"` to the command.

### Cursor

Add the server to `~/.cursor/mcp.json` (all projects) or `.cursor/mcp.json` in a project:

```
{
  "mcpServers": {
    "wbxsandbox": {
      "url": "https://sandbox.wbx.cz/mcp"
    }
  }
}
```

Cursor detects that the server needs a sign-in and opens our page in the browser. Check that the server is enabled under *Customize*; problems show in the Output panel under *MCP Logs*.

**Without a browser:** add `"headers": { "Authorization": "Bearer ${env:WBXSANDBOX_API_KEY}" }` and set the variable in your shell profile – Cursor reads `${env:…}` in remote server headers.

### VS Code (GitHub Copilot)

Run *MCP: Add Server* from the Command Palette, choose HTTP and paste the URL – or run *MCP: Open User Configuration* and add:

```
{
  "servers": {
    "wbxsandbox": {
      "type": "http",
      "url": "https://sandbox.wbx.cz/mcp"
    }
  }
}
```

VS Code opens the sign-in in the browser on first use. To check the connection, ask Copilot for something that needs a sandbox and expand the tool call – a plausible answer without a tool call means the server wasn't used.

**Without a browser:** add `"headers": { "Authorization": "Bearer ${input:wbx-key}" }` and an `inputs` entry `{ "type": "promptString", "id": "wbx-key", "description": "WBX Sandbox API key", "password": true }` – VS Code asks for the key once and stores it securely. Don't put the key itself in a shared `.vscode/mcp.json`.

### Codex CLI

```
codex mcp add wbxsandbox --url https://sandbox.wbx.cz/mcp
codex mcp login wbxsandbox
```

The configuration goes to `~/.codex/config.toml` and is shared with the Codex IDE extension. Check with `codex mcp list` or `/mcp` in the Codex terminal UI. On a machine without a browser, `codex mcp login wbxsandbox --no-browser` prints a URL to open elsewhere.

**With an API key instead:** in `config.toml` set `bearer_token_env_var = "WBXSANDBOX_API_KEY"` under `[mcp_servers.wbxsandbox]` – it names the environment variable, not the key itself.

### Gemini CLI

```
gemini mcp add --transport http wbxsandbox https://sandbox.wbx.cz/mcp
```

Then in Gemini CLI run `/mcp auth wbxsandbox` and sign in. `gemini mcp list` shows whether the server is connected.

**Known limitation:** the sign-in needs a local browser (Gemini CLI listens on `localhost` for the redirect), so it doesn't work over SSH or on a headless server. There, configure the server in `settings.json` with `"httpUrl": "https://sandbox.wbx.cz/mcp"` and `"headers": { "Authorization": "Bearer YOUR_API_KEY" }`.

### LangChain Deep Agents

No MCP needed – the `langchain-wbxsandbox` package gives your deep agent a sandbox as its execution backend. It is listed among the [sandbox integrations in the LangChain docs](https://docs.langchain.com/oss/python/integrations/sandboxes).

```
pip install langchain-wbxsandbox
export WBXSANDBOX_URL=https://sandbox.wbx.cz
export WBXSANDBOX_API_KEY=YOUR_API_KEY
```

Full example: [Examples → LangChain](https://sandbox.wbx.cz/docs/examples.md#langchain). Source: [github.com/WBX-Sandbox/langchain-wbxsandbox](https://github.com/WBX-Sandbox/langchain-wbxsandbox).

### Other MCP clients

Any client that supports remote MCP servers (Streamable HTTP) can use `https://sandbox.wbx.cz/mcp` with OAuth sign-in or with the header `Authorization: Bearer YOUR_API_KEY`. For clients that only start local programs (stdio), use our small helper program – downloads and configuration in [SDKs & MCP](https://sandbox.wbx.cz/docs/mcp.md).

> **What you connect to:** every sandbox is a separate Firecracker microVM with its own Linux kernel, on our own servers in the EU. You decide its network – full internet, only allowed destinations, or none. Background jobs come with an execution receipt (what ran, how, and what it cost), which WBX can sign, and prices are public on the [pricing page](https://sandbox.wbx.cz/pricing). Details in [Security and isolation](https://sandbox.wbx.cz/docs/security.md).

## If something doesn't work

- **The assistant doesn't see the tools** – Start a new chat and check the connector is switched on for it. In ChatGPT, Developer mode must stay on. In VS Code and Cursor, check the server is enabled and look at the MCP logs.
- **“Only team owners and admins can connect applications”** – Ask an owner of your WBX team to approve the connection, or to make you an admin. An owner or admin can also create an API key for you to use instead.
- **The sign-in page doesn't open or never returns** – The client waits for the redirect on `localhost` – it has to run on the same computer as your browser. Over SSH or on a server, use an API key instead (see the client's section above).
- **Do my files survive between chats?** – Yes, if you ask for a named sandbox (“use the WBX sandbox *my-project*”). Sandboxes started by your assistant fall asleep after 10 minutes without commands instead of being destroyed; the next command wakes them up with their files. Unused sleeping sandboxes are deleted after 7 days, or when you ask for it.
- **How do I disconnect?** – Delete the key named “… (connected app)” in *API keys* – it stops working with the next request. Or remove the server in the client.
- **What does it cost?** – Sandboxes and jobs the assistant runs are billed to your team like any other, per second. Set spending limits under *Billing*.

Source: https://sandbox.wbx.cz/docs/connect
