OAuth-protected AI tool access for IDE clients

The challenge

AI assistants such as IDE-based coding tools are most useful when they can reach an organization's internal tools: structured data, documents and application integrations. Exposing those tools safely requires one controlled entry point that knows who the caller is, what they are allowed to do and where to route each request.

What we built

The gateway validates identity-provider tokens, enforces group-based access control and routes tool calls to downstream MCP services. MonTek's founder owned the infrastructure path that makes it reachable and safe:

• Edge protection: Azure Front Door with web application firewall rules covering the MCP OAuth endpoints.
• Cluster entry: Kubernetes ingress on AKS for the gateway.
• Deployment: GitOps delivery with Flux, so changes are declared and reviewed in Git.
• Secrets: Key Vault wiring so credentials stay out of images and repositories.
• Public hostname: DNS and TLS termination for the public endpoint.

Diagnosing the client connection problem

While validating the path end to end, we traced connection failures between IDE-based MCP clients and the gateway to a protocol gap: the clients speak Streamable HTTP MCP (a POST to /mcp), while the gateway exposed a custom REST surface (/mcp/tools/*). We documented the remediation path for the application team so the gateway can speak the protocol that clients expect.

Results

• A public, TLS-terminated MCP entry point with OAuth metadata discovery and health probes.
• A repeatable pattern for onboarding additional identity-protected MCP services to AI coding assistants.
• A clear, documented fix for the protocol mismatch that was blocking clients.