Augment extension: add Sume as a remote MCP server over HTTP

In Augment's MCP settings, click + Add remote MCP, set the connection type to HTTP, name it sume, and enter https://mcp.sume.com/mcp as the URL. That is the whole form.

What the form does not show is a field for credentials. The Augment pages read for this post list URL and type only for remote servers, so authentication is the open question, and this post says how to find out in your install.

What are the exact fields?

The Augment MCP setup page lists three fields for a remote server: Connection Type (HTTP or SSE), Name, and URL. It also supports importing remote servers from JSON using url and type, where type is http or sse. For Sume:

{
  "mcpServers": {
    "sume": {
      "url": "https://mcp.sume.com/mcp",
      "type": "http"
    }
  }
}

Which type is right: HTTP or SSE?

HTTP. Sume's hosted endpoint is a remote HTTP MCP server at a single /mcp URL, per the MCP overview. Choosing SSE for it points the client at the wrong transport and shows up as a failed connection, not as an auth error.

How does Augment authenticate to Sume?

The extension page does not mention OAuth or headers. The Augment CLI page does show headers for HTTP and SSE entries when adding servers with auggie mcp add-json, and it warns that sensitive credentials should avoid direct configuration file storage.

Sume needs either an OAuth token or an API key, per OAuth and API keys. So the practical choice: in the extension, add the server and see whether it prompts for sign-in. If it does not and tools never appear, use the CLI route with a header instead. A connected Sume session proves itself when the agent can call mcp_health.

What can you do once connected?

Read-only discovery first: tools_list, catalog_list, jobs_list. Paid creation needs an idempotency_key per call and a session with write access; a read-only OAuth session returns insufficient_scope. Run dry_run=true on the first paid call to see the estimate before anything is spent.

What should you watch for with a stored key?

If you take the CLI header route, the Augment page warns that sensitive credentials should avoid direct configuration file storage. Follow that: load the key from an environment variable when the command runs, keep the file out of version control, and rotate the key if it ever appears in a log or chat.

Also keep one name per connection. Two entries pointing at the same Sume endpoint with different credentials will each show the same tool names, and an agent can pick the wrong one. A clear name such as sume-read and sume-write makes the permission obvious in every tool call.