AI Gateway lets you send traffic for every AI provider and MCP server to a single AI Gateway address. You tell AI Gateway which provider to route each request to using a request header, instead of relying on a separate endpoint per provider.
Routing requests before and after single-endpoint support
This example compares provider-specific routing with the single-endpoint model. Previously, each AI provider required its own endpoint path. With single-endpoint support, all requests can use the same base path, and the x-ns-aig-slug header identifies the provider or MCP server that AI Gateway should route the request to.
Before — a separate path per provider
curl https://<your-ai-gateway-fqdn>/v1/openai/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer <api-key>" \
-d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'
After — one path, header picks the provider
curl https://<your-ai-gateway-fqdn>/v1/chat/completions \
-H "Content-Type: application/json" \
-H "x-ns-aig-slug: openai" \
-H "Authorization: Bearer <api-key>" \
-d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'
To route the same request to a different provider, keep the destination unchanged and set x-ns-aig-slug to that provider’s slug instead. No other change needed.
The routing header
| Header name | Header value | Applies to |
|---|---|---|
x-ns-aig-slug | The provider’s slug — the same name shown in the AI Provider column (or the MCP tab) on the AI Gateway setup page | AI providers (LLM traffic) and MCP servers configured behind AI Gateway |
Find each provider or MCP server slug on the AI Gateway setup page in the Netskope UI; it matches the name shown there, so you do not need to create or look up a separate value. This applies whether the provider is predefined (for example, AWS Bedrock), a custom provider, or configured under the generic “Other” schema.
Using it with MCP servers
The single endpoint works the same way for Model Context Protocol (MCP) servers behind AI Gateway — send the request to the AI Gateway address and set x-ns-aig-slug to the MCP server’s slug, shown on the MCP tab of the AI Gateway setup page.
MCP servers that require OAuth
MCP clients that authenticate via OAuth cannot use header-based routing for the OAuth discovery flow, because discovery requests do not carry custom headers. If an MCP server requires OAuth, continue sending the full provider-specific destination URL for that connection. Header-based routing applies normally after a token is obtained.
MCP clients and duplicate server URLs
Some MCP client applications treat multiple servers with the same URL as a single connection and use only the first one, regardless of the header. Check your MCP client’s documentation to confirm it distinguishes servers by header before deploying multiple MCP servers behind the same address.