Connect a client

Connect an AI client

The console has a guided client setup that mirrors this page: pick your client, and it shows only the connection and auth options that client supports, then builds the exact config. This page is the reference for what each client needs and the steps to finish the connection.

Step 1

Pick your client

Start with the client. Each one supports a specific set of transports and auth methods, so choosing it first keeps you from configuring options that will not work.
ClientConnectionAuthWhere to set it up
ChatGPTAppsStreamable HTTP, SSEOAuthSettings > Apps > Advanced settings
Claude DesktopDesktopStreamable HTTP, SSE, LocalOAuthCustomize > Connectors
Claude.aiHostedStreamable HTTP, SSEOAuthCustomize > Connectors
MCP InspectorTestingStreamable HTTP, SSE, LocalBasic, Static, OAuthInspector transport selector
ManusCustom MCPStreamable HTTPStatic tokenSettings > Integrations > Custom MCP Servers
Other clientsManualStreamable HTTP, SSE, LocalBasic, Static, OAuthRemote MCP settings

Step 2

Choose a connection type

The transport the client uses to reach your endpoint. If you are unsure, choose Streamable HTTP.
TransportsStreamable HTTP recommended
  • Streamable HTTP: the current remote MCP transport. The default, and what most clients use.
  • SSE: a legacy event-stream transport, for clients that explicitly require it.
  • Local config: a local config file or launcher that points back to the hosted endpoint.

Step 3

Choose an auth method

How the client proves who it is. The console shows only the methods your chosen client supports.
CredentialsPick one
  • Static token: a bearer token. Tokens are stored hashed and the clear value is shown only once, right after you rotate it.
  • OAuth: the client runs its own OAuth sign-in flow. Best when a whole team needs access.
  • Basic login: your DataTether email as the username and your platform password.

Step 4

Copy the endpoint and connect

With the client, transport, and auth chosen, the console gives you the MCP URL and, where the client supports it, a ready-to-paste config.
  1. 1

    Open the client MCP or connector settings

    Client settings

    Go to the setup location for your client from the table above (for example, Customize > Connectors in Claude, or Settings > Apps in ChatGPT) and start adding a remote MCP server.

    What you'll seeAdd server

    You are adding a remote MCP server, named for your connection, that points at the hosted endpoint.

  2. 2

    Paste the MCP URL

    Console -> Connect

    Copy the endpoint URL from the console and paste it into the client. The URL is the same for every client; only the transport and auth differ.

    What you'll seeLive endpoint
    https://mcp.suriz.in/{org}/{system-name}/mcp

# {org} is your organization slug, {system-name} the connection's name.
  3. 3

    Set the auth, or import a config

    Choose the auth method and provide the credential: paste a static token, complete OAuth in the client, or enter your Basic login. For clients that accept a config file, import the generated config instead of filling fields by hand.

    What you'll seeConfig example
    {
  "mcpServers": {
    "your-system": {
      "type": "streamable-http",
      "url": "https://mcp.suriz.in/{org}/{system-name}/mcp",
      "auth": {
        "authentication": "bearer",
        "headers": { "Authorization": "Bearer YOUR_STATIC_TOKEN" }
      }
    }
  }
}
  4. 4

    Save and test by listing tools

    Save the connection, then ask the client to list available tools. If your governed tools appear, the client is connected and can start calling them under your access rules.

    What you'll seeConnected

    A working connection shows:

    • Your exposed tools in the client tool list
    • Read tools returning data, with writes still behind confirmation
    • The same endpoint reusable by any other MCP client

Related