# Connect your AI assistant

> Let Claude, ChatGPT, Cursor or any MCP-capable assistant run your hosting: sites, deploys, settings, domains, DNS and data.

Source: https://veldhost.eu/docs/connect-ai-assistant

veldhost has a built-in **MCP server**, the open standard AI assistants use to work with outside services. Connect it and your assistant can run your hosting with you: check health and read logs, deploy and approve releases, change a site's settings, connect a domain, edit DNS, look after backups and staging, query your data, and even [order a new site, email or a plan change for you to approve](https://veldhost.eu/docs/ai-ordering).

## The endpoint

```
https://manage.veldhost.eu/mcp
```

**The easy way: just sign in.** Add the URL above as a connector in your assistant and leave authentication empty. You'll be sent to the normal veldhost login (passkeys work), see exactly what the assistant will be allowed to do, and approve. You can revoke the connection any time under **Security → Connected apps**.

**Or use a token:** the same personal token as the [API](https://veldhost.eu/docs/api-reference), created under [API & MCP](https://manage.veldhost.eu/portal/api-tokens). Every client below has a copy-ready snippet on [Agents](https://manage.veldhost.eu/portal/agents) in the portal.

Either way, [permissions](https://veldhost.eu/docs/scopes-and-permissions) decide what the assistant may do, and it can never do more than your own role allows. With just **read** it can only look. Actions that cannot be undone (restoring a backup, enabling DNSSEC, dropping a warehouse database, approving or rolling back a release) always ask you first.

## Claude

In **Claude** (web, desktop and mobile): Settings → Connectors → *Add custom connector* → paste `https://manage.veldhost.eu/mcp`. No token needed: Claude walks you through the veldhost sign-in.

In **Claude Code**:

```
# OAuth (recommended: it opens the sign-in for you)
claude mcp add veldhost https://manage.veldhost.eu/mcp --transport http

# or with a personal token
claude mcp add veldhost https://manage.veldhost.eu/mcp \
  --transport http \
  --header "Authorization: Bearer mpc_..."
```

## ChatGPT

veldhost is **not yet listed** in ChatGPT's plugin directory, so you can't find it there by searching. Until it is, add it yourself with ChatGPT's **developer mode**. It works the same way, with the same veldhost sign-in. Developer mode is for Plus, Pro, Business, Enterprise and Education accounts on chatgpt.com in a browser. On a Business or Enterprise workspace, your admin may have to allow it first.

1. Turn on developer mode: in ChatGPT, open **Settings → Security and login** and switch on **Developer mode**.
2. Go to **Plugins** (`chatgpt.com/plugins`) and select the **+** button.
3. Name it `veldhost`. Under **Connection**, enter `https://manage.veldhost.eu/mcp` (with the `/mcp`), then create it. You don't need a token: ChatGPT finds our sign-in by itself.
4. Sign in to veldhost when ChatGPT asks, check what it will be allowed to do, and approve.
5. In a chat, open the **+** menu, choose **Developer mode** and select veldhost. Then ask, for example, *"Is my site up?"*

ChatGPT's menus change often. These steps were checked against OpenAI's own documentation in October 2026. If a name has moved, OpenAI's [developer mode guide](https://developers.openai.com/api/docs/guides/developer-mode) has the current path. In ChatGPT, veldhost offers the same tools as elsewhere, except the ones that price or order something (a new site, email, a plan change, staging, domains): ChatGPT sends you to the portal for those.

## Cursor, VS Code, Windsurf and others

**Cursor** (`~/.cursor/mcp.json`) and most other clients take this JSON; leave out `headers` to sign in with OAuth instead. **Windsurf** (`~/.codeium/windsurf/mcp_config.json`) takes the same, with `serverUrl` in place of `url`:

```
{
  "mcpServers": {
    "veldhost": {
      "url": "https://manage.veldhost.eu/mcp",
      "headers": { "Authorization": "Bearer mpc_..." }
    }
  }
}
```

**VS Code** (`.vscode/mcp.json`):

```
{
  "servers": {
    "veldhost": { "type": "http", "url": "https://manage.veldhost.eu/mcp" }
  }
}
```

**Gemini CLI** (`~/.gemini/settings.json`): add `"veldhost": {"httpUrl": "https://manage.veldhost.eu/mcp"}` under `mcpServers`. **Codex CLI**: `codex mcp add veldhost --url https://manage.veldhost.eu/mcp`. **OpenCode** (`opencode.json`): add `"veldhost": {"type": "remote", "url": "https://manage.veldhost.eu/mcp", "enabled": true}` under `mcp`; it opens the veldhost sign-in by itself, or run `opencode mcp auth veldhost`.

Any other client that speaks MCP over Streamable HTTP works too. For client developers: OAuth 2.1 with PKCE, discovery at `/.well-known/oauth-protected-resource/mcp`, dynamic client registration, and the MCP protocol versions 2026-07-28 down to 2024-11-05.

## What it can do

About ninety tools, grouped by the permission they need. The full, always-current list is the [MCP tool reference](https://veldhost.eu/docs/mcp-tools); in short:

- **Look** (read): sites, health, logs, deploys, traffic and analytics, backups, staging, cron, domains and registrar state, DNS, email, invoices, the audit log and the team.
- **Deploy**: deploy from git, set build steps, approve, discard or roll back CI/CD releases.
- **Manage**: environment variables, PHP or Node.js version, connecting a domain, staging, cron, backups and restores, webhooks.
- **DNS**: records, zone import and export, DNSSEC.
- **Data** (opt-in): read-only SQL on site databases, warehouse databases, dashboards, pipelines and jobs.
- **Billing** (opt-in): propose orders for you to approve, see [Ordering with an AI assistant](https://veldhost.eu/docs/ai-ordering).

Paying, deleting a site or a domain, changing nameservers, creating tokens and changing the team stay in the portal, signed in as you; the assistant sends you a link.

## Tips for the best results

- Start with *"What's on my veldhost account?"*: the assistant reads `veldhost://account` and knows your sites, domains and permissions.
- For a broken site, ask it to *diagnose* it: the `diagnose_site` prompt walks health, deploys, logs and traffic in order.
- Everything the assistant does is in your **audit log**, with the app that did it.