Apigee hybrid adds MCP support for on-prem agents

What this means for multi-cloud and on-premises teams Agentic endpoints run where your APIs already run. Apigee hybrid already lets you host the runtime on the Kubernetes platform of your choice. MCP inherits this capability. The MCP sidecar reuses the existing apigee-watcher identity. There is no new Google service account to create. There is no new IAM role to grant. One control-plane grant per organization handles the rest. The identity you already have does the rest of the work.

Data residency and sovereignty posture carry over. MCP in Apigee hybrid is available for Subscription and Evaluation organizations. This includes Data Residency–enabled organizations. These organizations are configured at organization setup. On hybrid, configuration is fetched from your control plane location. The runtime request path stays in your cluster by construction. For teams with strict constraints, the architecture supports them. The request may not leave this jurisdiction, this facility, or this VPC. That constraint is supported by the architecture rather than by a setting.

The governance you already wrote applies to agents. MCP endpoints are Apigee proxies. They inherit the platform features you know. These include OAuth 2.0 and OpenID Connect. They include API key verification. They include Apigee's built-in policy library. Tool-level control comes from a concept you already use. This concept is the API product. You list the specific tool operations a client is entitled to. You can optionally set a per-tool quota. Apigee enforces these quotas automatically. It also filters discovery automatically. When a client calls tools/list, the response contains only the tools in that client's API product. An agent cannot call delete_account if it was never allowed to see it.

Your agentic traffic shows up where your API traffic does. Deployed MCP Discovery Proxies are automatically ingested into Apigee API hub. They are assigned the MCP API style. Their operations map to individual tools. This creates a searchable catalog that developers can query. Developers can query this catalog in natural language. They do not need a spreadsheet of endpoints. In API insights, two filters isolate agentic traffic from ordinary REST traffic. These filters are API style and MCP tool name. You can look at a single tool across the API performance dashboard. You can see error analysis, latency, and target performance. You can see the ratio of discovery to execution. You can tell proxy overhead apart from backend time.

MCP is an optional feature in Apigee hybrid. It is not enabled by default. Enabling it is a Helm operation on a cluster. The cluster must already run Apigee hybrid 1.17.0 or later. Steps to do this are as follows. Set the top-level enableMcpServer: true key in your overrides.yaml. Then upgrade the apigee-operator chart first. Then upgrade the apigee-org chart second. The order matters for this operation. The operator chart owns the schema for the new MCP resources. Upgrading in the wrong order produces a successful upgrade with no MCP pods. There is also a one-time, per-organization grant. This grant adds the apigee-watcher identity to your organization's control plane access. This allows the sidecar to retrieve configuration. Defaults are sensible out of the box. The setup includes two MCP replicas. It supports autoscaling to ten. Rolling back is the same field set to false. You perform a single chart upgrade to roll back.

A few things to plan for before you schedule the change. Enabling or disabling MCP triggers a staged canary release of runtime pods. You must run this in an approved maintenance window. The public documentation describes how to enable MCP in an existing Apigee hybrid cluster.

Your MCP proxy is exposed as a single endpoint. Every agent request arrives at the same /mcp path. Apigee tags each of those requests with the name of the tool that was called. API insights can break that one endpoint out by tool. This shows which tools your agents actually used. It shows how often they were used. It shows usage over what period. The screenshot below shows API insights, including requests to the MCP proxy. It also shows individual tools as API proxies. Examples include mcp-healthcare, mcp-petstore, and mcp-statuscode. You can see response time and tps for these tools. You can see the public documentation on monitoring and analyzing mcp traffic with Apigee API insights. You can also see how to use the API insights dashboard.

We also have a view of tools being called in the tool performance tab within API insights. This view provides detailed metrics on tool usage.

What's next: tracing an agent's call end to end An MCP tool call crosses an AI client, Apigee, and your backend. Today each of those components is its own island of logs. The question platform teams actually ask is where did the time go. This question has no single answer today. The snippet below is an example mcp tools call request from an AI client. The curl request below allows for end to end tracking of the request. This tracking works when distributed trace is enabled in Apigee with OTEL.

The same call as orchestrated by distributed trace in Google CloudTrace shows the calls in its entirety. This view provides one global view of the call. The assumption is that all services in the chain report to the same Google CloudTrace. This includes Apigee and non-Apigee services. MCP clients carry W3C trace context inside the JSON-RPC _meta field. A conventional REST backend has no idea how to read this field. Apigee extracts it and forwards it as ordinary traceparent and tracestate headers. Instrumented services downstream join the trace without code changes. Three things follow from this setup. A single trace ID spans the AI client, the Apigee proxies, and backend execution. Backends that participate do not need recoding. The gateway can forward context in passthrough mode. This reduces noise and trims trace ingestion cost.

The ecosystem is converging on the same idea. SEP-2028, an open proposal in the MCP community, aims to standardize _meta-to-header forwarding of W3C trace context. This standardizes standard behavior for MCP servers. The OpenTelemetry path that carries W3C trace context is available in Apigee today.

MCP support is available now in Apigee hybrid 1.17.0. Start with the MCP in Apigee overview for the architecture. Use Enable MCP in Apigee hybrid for the Helm procedure. Use the MCP quickstart to get a first tool answering an agent.

Why it matters Multi-cloud teams can keep data sovereignty without moving their runtime.

How OpenSmartRoute helps A team that routes its requests through OpenSmartRoute gains from this news by securing its on-premises agent traffic. OpenSmartRoute sends each request to the best-fit model, agent, tool, or skill from a catalogue the team defines. It scores every candidate on quality, cost, speed, and safety. The team sets the weights per request. Hard rules pin a request: text with personal data stays on an on-premises model. A region or a cost cap is never crossed. It learns from outcomes, so a model that answers well gets more traffic. One model that fails gets less traffic. A new model is one catalogue entry and competes on the next request. Nothing else changes in the app. It works with any OpenAI-compatible provider, with open-weight models served locally, with MCP tools and A2A agents. A savings ledger shows what each routed request cost next to what the most expensive model would have cost. An input guard spots prompt injection and personal data before a request leaves. 'osr eval' measures routing accuracy on the team's own prompts and can fail a build when it drops. The hosted platform keeps a models catalogue with prices and public rankings built from real traffic.

To set this up, the team would define a catalogue entry for the Apigee MCP proxy. The entry would specify the endpoint URL and the MCP tool capabilities. The team would add a rule in OpenSmartRoute to route requests containing PII to the on-premises model. This ensures data never leaves the cluster. The team would enable 'osr eval' to test the routing accuracy against their own prompts. The savings ledger would show the cost reduction compared to sending all traffic to a public cloud model.

What to do Enable MCP in your Apigee hybrid cluster using the Helm procedure. Set the enableMcpServer key to true in your overrides.yaml. Upgrade the apigee-operator chart first, then the apigee-org chart second. Run the change in an approved maintenance window to handle the staged canary release. Verify that the MCP proxy is exposed at the /mcp path. Check API insights to confirm that tool calls are tagged correctly. Monitor the tool performance tab to ensure agents are discovering and calling the right tools. Review the security logs to confirm that no traffic crosses the boundary to Google Cloud.

How OpenSmartRoute helps

A team routing requests through OpenSmartRoute gains immediate visibility into how Apigee hybrid 1.17.0 handles new MCP traffic. The router can now score every candidate model on quality, cost, speed, and safety using the updated Apigee environment. This allows the team to set weights for each metric and hard rules for data privacy within the cluster.

The setup requires adding Apigee hybrid 1.17.0 to the models catalogue with its specific pricing and latency figures. OpenSmartRoute will route text containing personal data to on-premises models, ensuring no data leaves the Kubernetes cluster. The input guard will spot prompt injection attempts before any request reaches the Apigee proxy.

A savings ledger will show exactly how much each routed request cost compared to the most expensive model. The 'osr eval' tool can measure routing accuracy on the team's own prompts to ensure the new MCP support does not degrade performance. If evaluation scores drop, the build fails automatically, preventing bad routing from going live.

Source: https://developers.googleblog.com/announcing-mcp-support-in-apigee-hybrid-agentic-tools-that-run-in-your-own-cluster/