Hi folks,
We’re designing an MCP integration where each customer is served by a fully isolated, tenant-specific MCP server, rather than a single centralized MCP service with shared OAuth.
Concretely, our model looks like this:
-
Each tenant has its own MCP endpoint, for example:
-
https://f5.company.com/mcp -
https://discord.company.com/mcp
-
-
No central OAuth authority across tenants.
-
Authentication and authorization are handled per-tenant, scoped to that domain.
-
The MCP server trusts requests only within its own tenant boundary (zero-trust across tenants).
-
Domain ownership and verification are per-tenant, not global.
We wanted to confirm a few things about ChatGPT’s MCP support:
-
Does ChatGPT support discovering and connecting to multiple tenant-scoped MCP servers under different subdomains for the same product?
-
Is a centralized OAuth or identity provider required for official MCP connectors, or are per-tenant auth flows acceptable?
-
For official app / connector verification:
-
Can verification tokens and
.well-knownendpoints be configured independently per tenant domain? -
Are there any constraints that assume a single global MCP endpoint?
-
-
Are there recommended patterns or known limitations when implementing a zero-trust, multi-tenant MCP architecture like this?
We’re aiming to avoid cross-tenant coupling entirely and want each customer to remain isolated at the protocol, auth, and infrastructure levels.
Would love guidance on whether this architecture is supported today, or if there are roadmap considerations we should be aware of.
Thanks in advance.