Connect other AI clients
Which AI clients can connect to the SkaleData MCP gateway, how each one signs in or uses an API key, and what to do when a client is refused.
The SkaleData MCP gateway is one HTTP endpoint that AI clients use to work with your clusters, apps, Superset, Airflow, Airbyte, and DataHub. Connect Claude chat to SkaleData describes the tools and settings in detail, and What data leaves your cloud describes what reaches your AI provider and what SkaleData records. This page is about the clients themselves: which ones can connect today, and how.
https://mcp.skaledata.run/mcpBefore you begin
- An org admin has turned on MCP data egress for your organization. While it's off, the gateway refuses every tool call from every client. Read What data leaves your cloud first.
- You've signed in to the SkaleData console at least once with the account you'll use, and you're a member of the organization you'll choose.
Two ways to connect
| Sign in as yourself | API key | |
|---|---|---|
| How | The client opens a browser; you sign in with your SkaleData account and choose an organization. | The client sends an org-wide API key with the full scope in an Authorization header. |
| Identity | You, with your own roles and workspace grants. | A shared SkaleData service identity. |
| Tools | Every tool, including Airflow, Airbyte, and DataHub, which act as you. | Clusters, apps, health, deploys, and Superset. The Airflow, Airbyte, and DataHub tools refuse keys, because they act as a person. |
| For | People using an AI client. | Scripts, CI jobs, and clients that can't sign in. See Connect other MCP clients with an API key. |
Sign in when you can. The gateway identifies a client that signs in by its OAuth client identity, and allows a fixed list of clients; the table below says where each one stands.
Clients
| Client | How it connects today |
|---|---|
| Claude chat | Signs in. Add SkaleData as a custom connector: Connect Claude chat to SkaleData. |
| Claude Code | Signs in. Install the SkaleData plugin, then /mcp › Authenticate. Without the plugin, from a shell: claude mcp add --transport http skaledata https://mcp.skaledata.run/mcp, then claude mcp login skaledata (or, inside Claude Code, /mcp › Authenticate). |
| VS Code with GitHub Copilot | Signs in. Add the gateway as an HTTP MCP server (below); VS Code opens a browser for the sign-in on the first connection. |
| ChatGPT | Signs in. Add the gateway as a custom MCP server (below). |
| Cursor | Signs in using the configuration below, which includes SkaleData’s public Cursor client ID. |
| Codex | Signs in. Add the gateway to your Codex configuration, then run codex mcp login (below). |
| Any other MCP client | API key, as in Connect other MCP clients with an API key. A client that publishes its identity as a Client ID Metadata Document can sign in once SkaleData has allowed it; see Ask for a client to be allowed. |
VS Code with GitHub Copilot
- Open the Command Palette and run MCP: Add Server.
- Choose HTTP (HTTP or Server-Sent Events), enter
https://mcp.skaledata.run/mcp, and give the server an ID such asskaledata. - Choose where to save it. For a server you use in every workspace, choose the global Copilot configuration.
The saved entry looks like this, in VS Code's servers format:
{
"servers": {
"skaledata": {
"type": "http",
"url": "https://mcp.skaledata.run/mcp"
}
}
}On the first connection, VS Code opens a browser window where you sign in to SkaleData and choose an organization. You don't need an oauth block or a client ID: VS Code identifies itself to the gateway on its own. See MCP servers in VS Code and the MCP configuration reference.
ChatGPT
- Open ChatGPT Plugins at chatgpt.com/plugins, select the plus button, and then Add custom MCP server.
- Enter a name, such as
SkaleData, and a description. - Under Connection, choose Public endpoint and enter
https://mcp.skaledata.run/mcp. - Configure authentication to sign in, read the risk warning, select I understand and want to continue, and then Create as a plugin.
- Sign in to SkaleData in the window that opens and choose an organization. ChatGPT then shows the tools it found.
Your account or workspace policies decide whether you can add custom MCP servers; a workspace admin may have to allow it. See Connect your MCP server to ChatGPT. If ChatGPT says authentication succeeded but discovery failed, see Troubleshooting.
Cursor
Add this entry to ~/.cursor/mcp.json for all workspaces, or .cursor/mcp.json in one project:
{
"mcpServers": {
"skaledata": {
"url": "https://mcp.skaledata.run/mcp",
"auth": {
"CLIENT_ID": "https://docs.skaledata.com/oauth/cursor-client.json",
"scopes": ["user:org:read", "skaledata:mcp", "offline_access"]
}
}
}
}Enable the server in Cursor's MCP settings and connect. Sign in to SkaleData in the browser and choose your organization. The consent screen identifies this connection as Cursor (SkaleData). You don't need an API key or client secret.
For Cursor Agent in a terminal, use the same configuration and run:
cursor agent mcp enable skaledata
cursor agent mcp login skaledataIf you previously configured an API key for this server, remove its Authorization header to use sign-in. Keep the CLIENT_ID above: a URL-only Cursor entry attempts dynamic registration, which this gateway does not support. See Cursor's MCP configuration.
Codex
With a current Codex CLI, add this server to ~/.codex/config.toml:
[mcp_servers.skaledata]
url = "https://mcp.skaledata.run/mcp"
scopes = ["user:org:read", "skaledata:mcp", "offline_access"]Then sign in:
codex mcp login skaledataComplete the browser sign-in and choose your SkaleData organization. Use the same login command to reconnect later. The offline_access scope lets Codex renew the connection when its access token expires.
Codex supplies its own public client identity. No client ID, client secret, or API key is needed in your configuration. If you already have a bearer-token entry for this gateway, replace it with the entry above to sign in as yourself. See MCP in Codex.
Cursor and Codex can also use an API key as described in Connect other MCP clients with an API key. That connection acts as a shared service identity; Airflow, Airbyte, and DataHub tools require sign-in as a person.
Ask for a client to be allowed
The gateway allows a fixed list of clients to sign in. As of October 2026, that list is Claude chat, Claude Code, VS Code, ChatGPT, Cursor, and Codex. A client that isn't on the list completes the sign-in and is then refused with client not allowed.
To ask for a client, tell SkaleData the client’s name. It needs a public client identity that the gateway can allow, either supplied automatically or configured explicitly as in Cursor’s setup above. Until the client is supported, it can use an API key if it can send an Authorization header.
Choose which clients your organization allows
By default, people in your organization can sign in from every client SkaleData supports. An org admin can limit that to a chosen few:
- In the SkaleData console, go to Organization > General.
- In AI clients, choose Only these, select the clients to allow, and save.
A person who signs in from a client you didn't select is refused with "Your organization hasn't allowed … yet. Ask an org admin to allow it under Organization › General." Once saved, clearing a client takes effect on that person's next request. The choice applies to sign-ins only; API keys stay governed by their scopes and by MCP data egress, as described in What data leaves your cloud.
Troubleshooting
| What you see | What it means | What to do |
|---|---|---|
| "client not allowed", or the sign-in succeeds but the client finds no tools | The client isn't on the gateway's list of allowed clients yet. | Ask SkaleData to allow the client, naming it. Until then, use an API key if the client can send a header. |
| "Your organization hasn't allowed … yet" | SkaleData supports the client, but your organization limits which clients people may sign in from, and this one isn't selected. | Ask an org admin to select it under Organization > General > AI clients. See Choose which clients your organization allows. |
| "MCP data egress is disabled for this organization" | Your organization hasn't allowed data to leave for AI clients. | Ask an org admin to turn on Allow MCP data egress in the SkaleData console, under Organization > General. |
| The client says the server needs authentication, or a 401 | The client has no valid sign-in for the gateway, for example after it expired or you signed out. | Sign in again from the client: /mcp › Authenticate in Claude Code, cursor agent mcp login skaledata in Cursor Agent, codex mcp login skaledata in Codex, or reconnect in the client’s settings. |
| "Airflow tools need a person signed in through an AI client such as Claude", or the same for Airbyte or DataHub | The connection uses an API key, and these tools act as a person. | Use a client that signs in, from the table above. |
| A generic error right after you sign in | SkaleData refused the sign-in, usually because you haven't signed in to the console with this account yet, or you aren't a member of the organization you chose. | Sign in to the SkaleData console once with the same account, check your membership, and connect again. |
Related pages
Install the Claude Code plugin
Install SkaleData's plugin for Claude Code, sign in as yourself, and use the SkaleData MCP gateway and the skills for diagnosing a failed Airflow run and checking the Airbyte sync behind a stale table.
Connect other MCP clients with an API key
Connect a script, a CI job, or another MCP client to the SkaleData MCP gateway with an org-wide API key, and what changes when the client acts as a key instead of a person.