I already have a DataUnmess MCP key. Install the DataUnmess MCP server into this AI client.

Use this endpoint: https://app.dataunmess.ai/api/mcp

First detect which client/environment you are running in.

If this is Claude Code (the CLI / VS Code extension), do NOT hand-edit ~/.claude.json. Run this command exactly — the name and URL are positional and must come before the flags:

claude mcp add dataunmess https://app.dataunmess.ai/api/mcp --transport http --scope user --header "Authorization: Bearer <YOUR_MCP_KEY>"

If this is Codex Desktop, use the TOML config below. If this is Claude Desktop, Cursor, Windsurf, ChatGPT, Gemini, or another remote-MCP JSON client, use that client's native MCP JSON config. If you are unsure which config file to edit, fetch https://app.dataunmess.ai/connect-mcp for the current snippets or ask me which client I am using before editing.

For Codex Desktop, the config block is:

[mcp_servers.dataunmess]
url = "https://app.dataunmess.ai/api/mcp"
http_headers = { Authorization = "Bearer <YOUR_MCP_KEY>" }

For JSON-style remote MCP clients, the config entry is:

{
  "mcpServers": {
    "dataunmess": {
      "url": "https://app.dataunmess.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_MCP_KEY>"
      }
    }
  }
}

Keep the literal string <YOUR_MCP_KEY> as the bearer token. Keeping the real key out of this chat is intentional; I will paste it into the local config file (or the `claude mcp add` command) myself after you save it.

After saving, remind me to replace <YOUR_MCP_KEY> with my real key, restart or reload the AI client, wait up to 60 seconds if the MCP server list does not refresh immediately, then test with: "Start DataUnmess Memory onboarding. Ask me for my company website, research public pages, and seed DataUnmess Memory markdown with company context, products, glossary, standard KPIs, and useful business terms. Then propose three dashboards we should create first.".

If the AI can start that onboarding flow with DataUnmess MCP tools, the connection is working.