Connect an app

Claude app and Desktop

Opens Claude's add-connector dialog with the name and URL filled in. You read the values, press Add, then press Connect and sign in.

On Team and Enterprise an Owner adds it once for everybody, and each member then signs in as themselves.

Sign in to Claude Code with the same claude.ai account and it picks this connector up on its own. Nothing else to run.

Claude Code

Add the server from a shell, then sign in. Leave the Authorization header off if you want to sign in rather than hold a token.

claude mcp add --transport http splitscreen https://splitscreen.chat/api/mcp

Then run /mcp in a session, or claude mcp login splitscreen from the shell, and finish in the browser. The login command needs Claude Code 2.1.186 or newer; /mcp works on any version.

The download is .mcp.json. Commit it at the root of a repo and your whole team gets the server; Claude Code asks each of them to approve it once, when they open an interactive session there. A claude -p run, an Agent SDK session or a cloud session loads it without asking.

{
  "mcpServers": {
    "splitscreen": {
      "type": "http",
      "url": "https://splitscreen.chat/api/mcp"
    }
  }
}

ChatGPT

Settings, Security and login, Developer mode; then Plugins, the plus button, a name and description, paste the URL, choose OAuth, create.

ChatGPT has no prefilled link, so the URL is typed or pasted. Developer mode is Plus and up, on the web; a free account cannot add a custom MCP server at all. On Business and Enterprise an admin allows it before a member can.

OpenAI runs a directory that ChatGPT and Codex share, and a listed plugin installs from it with one press and no developer mode. Ours is not submitted yet. When it is, this row gets shorter.

Cursor

Cursor asks you to confirm before it adds anything. The dialog should show the URL below and no command.

The encoded half of that link is {"url":"https://splitscreen.chat/api/mcp"}. If the button does nothing, which is what a managed machine, WSL or a remote session tends to do with a cursor:// link, that object is the entry for splitscreen under mcpServers in .cursor/mcp.json.

VS Code

VS Code asks you to trust the server before it starts it. The dialog should show the URL below and no command.

The encoded half of that link is {"name":"splitscreen","type":"http","url":"https://splitscreen.chat/api/mcp"}. Where a vscode: link goes nowhere, on a managed machine or over SSH, run MCP: Add Server from the command palette and paste the URL, or put that same object under servers in .vscode/mcp.json.

Gemini CLI

No link exists for this one. Put the snippet in ~/.gemini/settings.json and run /mcp auth splitscreen.

{
  "mcpServers": {
    "splitscreen": {
      "httpUrl": "https://splitscreen.chat/api/mcp",
      "timeout": 300000
    }
  }
}

Or add it from a shell, rather than editing the file:

gemini mcp add --transport http splitscreen https://splitscreen.chat/api/mcp

httpUrl is Streamable HTTP. url is the older SSE transport and this endpoint does not serve it.