Grafana Cloud MCP: OAuth refresh succeeds but SSE stream fails with Bad Request (reconnect loop)

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

The Grafana Cloud marketplace MCP (plugin-grafana-cloud-mcp-grafana-cloud) connects initially via streamableHttp, then fails on SSE reconnect with HTTP 400 Bad Request. OAuth token refresh completes successfully, but the transport stays broken and enters a reconnect loop until maximum attempts are exceeded.

Steps to Reproduce

  1. Enable Grafana Cloud MCP plugin in Cursor and authenticate via OAuth.\n2. Observe initial connect_success in MCP Logs.\n3. Leave session idle or continue agent work for 30+ minutes.\n4. Watch MCP Logs for: Failed to open SSE stream: Bad Request; Maximum reconnection attempts exceeded.

Expected Behavior

Stable streamable HTTP / SSE transport after OAuth refresh, or graceful reconnect without permanent failure.

Operating System

MacOS

Version Information

Cursor IDE (macOS). Grafana Cloud MCP plugin (streamable HTTP). MCP Logs 2026-08-26.

Additional Information

macOS darwin 25.6.0. Contributes to parallel createClient timeouts on other MCP servers. Related: MCPService: Timeout waiting for EverythingProvider (mcp.createClient) while stdio MCP server stays healthy Workaround: disable Grafana Cloud MCP when not in use.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Hey @adiedx-alkami
To get it working again without leaving it off: toggle the Grafana Cloud plugin off and back on, or run Reload Window from the Command Palette - either one forces a fresh session and reconnects it. It’s an issue we’re tracking on the streamable-HTTP transport side; I’ll follow up here when there’s an update.