# Connect a client

Add the Kelrova MCP server to Claude Code, Claude, ChatGPT, Codex, Cursor or VS Code, with the configuration each one needs.

*Where in the portal: Organization → MCP server → Connect a client*

*Last updated: 2026-10-21*

The server address is `https://mcp.kelrova.com/mcp`. Enter it exactly like this wherever a client asks for it: some clients refuse a URL that differs from the one Kelrova publishes. Each client opens your browser once for consent; see [Access levels and staying connected](https://kelrova.com/docs/mcp/access).

![The steps for connecting one editor to the Kelrova MCP server: the server address and the setup it needs.](https://kelrova.com/docs/img/mcp/connect.png)

*Connect a client: the server URL and the steps for one client.*

> **Note: Checked against vendor documentation**
>
> Kelrova has not yet run these clients against its server. These steps are what each vendor documents (checked 2026-10-16). Where a detail is not documented, the step says it is unconfirmed: it may work, but it is not promised. A shorter version of these steps is under **Connect a client** on the **MCP server** page.

## Choose your client

### Claude Code

Add the server:

```bash
claude mcp add --transport http kelrova https://mcp.kelrova.com/mcp
```

Then, in a session, run `/mcp`, pick **kelrova** and choose **Authenticate**. A browser window opens for consent. To sign in from a shell instead, run `claude mcp login kelrova`. `claude -p` cannot run the sign-in, so sign in interactively once first.

**Pin the access you want.** When a tool needs more access than the connection has, Claude Code asks you to sign in again, but it requests what its own `oauth.scopes` setting lists, not what Kelrova names in its response. Without the pin, `request_recheck` sends you through sign-in again and again without ever receiving the re-check access. Include `offline_access` so the connection stays alive without signing in again:

```bash
claude mcp add-json kelrova '{"type":"http","url":"https://mcp.kelrova.com/mcp","oauth":{"scopes":"findings:read findings:write findings:recheck offline_access"}}'
```

To share the setup with a team, commit it in the project's `.mcp.json`. An entry with a `url` and no `type` is rejected by Claude Code.

**.mcp.json**

```json
{
  "mcpServers": {
    "kelrova": {
      "type": "http",
      "url": "https://mcp.kelrova.com/mcp",
      "oauth": {
        "scopes": "findings:read findings:write findings:recheck offline_access"
      }
    }
  }
}
```

Each person still approves their own consent screen, and can still grant viewing only there. How Claude Code behaves when it asks again and you grant less than it asked for is unconfirmed.

### Claude

For Claude on the web, the desktop app and mobile:

- **Free, Pro and Max:** open **Customize → Connectors → Add custom connector**. Name it `Kelrova` and enter `https://mcp.kelrova.com/mcp`.
- **Team and Enterprise:** an owner opens **Organization settings → Connectors → Add → Custom** and enters the URL. Members then open **Customize → Connectors**, find the connector with the **Custom** label and choose **Connect**.
- Choose **Sign in now**. For the OAuth client keep **Use Claude's published identity** (recommended); **Register automatically** also works. Leave the client ID and secret empty.
- Authentication settings cannot be edited after the connector is added: remove it and add it again.

The dialog layout differs by organization. Claude documents that it asks you to approve more access when a tool needs it; this is unconfirmed with Kelrova.

### ChatGPT

- Open `chatgpt.com/plugins`, choose **+** and **Add custom MCP server**.
- Enter a name and description, choose the public endpoint `https://mcp.kelrova.com/mcp`, set authentication to OAuth (the client identity published by ChatGPT, or automatic registration), accept the warning and choose **Create as a plugin**.
- Install it and select it with `@` in a new conversation.
- Workspace policies decide who may add custom servers. Which ChatGPT plans and roles can add one is unconfirmed.

ChatGPT documents its own way of asking for additional access; whether that works with Kelrova is unconfirmed. Use the **Re-check** button in the portal rather than counting on `request_recheck` from ChatGPT.

### Codex

Add the server and sign in:

```bash
codex mcp add kelrova --url https://mcp.kelrova.com/mcp
codex mcp login kelrova
```

Or add it to your Codex configuration file:

```toml
[mcp_servers.kelrova]
url = "https://mcp.kelrova.com/mcp"
```

Codex chooses between the published client identity and automatic registration on its own. To force the published identity for one login, add `--oauth-client-registration cimd`.

Unconfirmed: which access Codex asks for at first sign-in (you can grant viewing only on the consent screen), how it refreshes a connection and whether it asks for more access later.

### Cursor

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

```json
{
  "mcpServers": {
    "kelrova": {
      "url": "https://mcp.kelrova.com/mcp"
    }
  }
}
```

Cursor starts the sign-in when you enable the server. It documents automatic registration only; the client identity method Kelrova prefers is unconfirmed for Cursor. Organizations that restrict connections to a list of apps refuse automatic registration. In that case ask an owner or administrator to allow the client; see [Organization settings for MCP](https://kelrova.com/docs/mcp/organization).

Unconfirmed: refresh, and asking for more access later.

### VS Code

For VS Code with Copilot, add the workspace file `.vscode/mcp.json`:

```json
{
  "servers": {
    "kelrova": {
      "type": "http",
      "url": "https://mcp.kelrova.com/mcp"
    }
  }
}
```

Or run **MCP: Add Server** from the command palette, choose **HTTP**, and enter the URL and a server ID. A browser window opens on first use.

Unconfirmed: a command-line form for adding an HTTP server, refresh behavior and asking for more access later.

## After you connect

The connection appears under **Settings → Connected apps**. Ask your agent to use the `fix_security_findings` prompt, or follow [A typical run](https://kelrova.com/docs/mcp/loop). If the connection fails, see [Troubleshooting](https://kelrova.com/docs/mcp/troubleshooting).

## Related

- [Access levels and staying connected](https://kelrova.com/docs/mcp/access)
- [The six tools](https://kelrova.com/docs/mcp/tools)
- [Organization settings for MCP](https://kelrova.com/docs/mcp/organization)
- [Troubleshooting](https://kelrova.com/docs/mcp/troubleshooting)