1. Add MCP Servers to the Hub

MCP servers connect to the private hub through the UI or API. The hub enables shared access across workspaces and team members. Portkey supports remote MCP servers via StreamableHTTP and SSE transports. StreamableHTTP servers are recommended—SSE will be deprecated in the future.
  1. Add the complete server URL (e.g., https://mcp.linear.app/mcp)
  2. Set the transport type. When uncertain, select Streamable HTTP
  3. Choose the auth type expected by the MCP server:
    • OAuth: Portkey automatically handles dynamic client registration with authorization code flows
    • Client Credentials
    • Custom Headers
  4. Set auth granularity: org-wide, workspace-level, or user-level
  5. Add the server

2. Provision Access to Workspaces

MCP servers become available for sharing with any workspace in the account. Workspace managers can also initiate server addition directly from within their workspace. Portkey generates a URL for each MCP server, accessible by workspace members:
https://mcp.portkey.ai/:workspace-id/:mcp-server-id/mcp
For self-hosted MCP gateways:
https://mcp-gateway-url/:workspace-id/:mcp-server-id/mcp

3. Use MCP Servers in Apps and Agents

Integrate these MCP server URLs in your applications and agent workflows. For example, add it to cursor at Cursor Settings > MCP & Integrations > New MCP Server The final configuration would look like this
{
	{
		"mcpServers": {
			"server1": {
				"url": "https://mcp.portkey.ai/:workspace-id/:mcp-server-id/mcp"
			},
			"server2": {
				"url": "https://mcp.portkey.ai/:workspace-id/:mcp-server-id/mcp"
			}
		}
	}
}
Note that Portkey manages auth end to end. If the upstream MCP server needs auth, Portkey will prompt the user to allow access to the upstream server and then complete the auth flow with that server. This keep the developer experience extremely clean while maintaining granular access control and auth.

Next

  • View logs & traces for MCP interactions
  • Supported capabilities
  • Authentication deep dive
  • Add guardrails