Apigee Hybrid Adds MCP Support for Local Agentic API Tooling
Run Model Context Protocol data planes directly within your own Kubernetes clusters.
1Local MCP Data Plane Deployment
Apigee hybrid 1.17.0 introduces Model Context Protocol support, allowing the data plane to run inside your own Kubernetes cluster. This architecture addresses critical requirements for customers who must keep runtime operations within their own data centers or private clouds due to strict regulatory mandates or specific multi-cloud business infrastructure realities.
2Simplified Agentic Tool Integration
Developers can now transform existing APIs into AI-discoverable tools without writing custom MCP server code. By using an OpenAPI 3.0.x specification, Apigee generates an MCP Discovery Proxy. Agents interact via JSON-RPC, utilizing standard tools/list and tools/call methods while Apigee handles all necessary protocol transcoding and lifecycle management automatically.
3Security and Traffic Isolation
Security teams benefit from a design where MCP request traffic never leaves the local cluster. All authorization and proxying to backends occur entirely within your infrastructure. Furthermore, configuration is pulled by a sidecar, ensuring the Google Cloud control plane never initiates inbound connections into your private network, maintaining strict perimeter security.
4Helm-Based Configuration Process
Enabling MCP is an optional Helm operation requiring version 1.17.0 or later. Users must set the enableMcpServer key to true in their overrides.yaml and upgrade the apigee-operator chart before the apigee-org chart. This specific sequence is mandatory to ensure proper resource schema initialization and prevent the failure of MCP pod creation.
5Operational Canary Release Strategy
Enabling or disabling MCP triggers a staged canary release of runtime pods across the environment. Organizations should execute these changes during approved maintenance windows to mitigate potential service disruptions. The system defaults to two MCP replicas with autoscaling capabilities up to ten, providing a robust starting point for production workloads.
6Granular API Insights Monitoring
Every agent request arrives at a single /mcp endpoint, which Apigee tags with the specific tool name invoked. This allows API insights to break down usage metrics by individual tools, such as mcp-healthcare or mcp-petstore. Teams can monitor response times and throughput per tool to optimize agentic performance and identify usage patterns.
7Distributed Tracing and Observability
Apigee extracts W3C trace context from the JSON-RPC _meta field and forwards it as standard headers to downstream backends. This enables a single trace ID spanning the AI client, the gateway, and the backend. Instrumented services join the trace without code changes, significantly reducing noise and lowering overall trace ingestion costs.
8Standardization and Future Outlook
The industry is converging on standardized _meta-to-header forwarding, as seen in the SEP-2028 open proposal. By leveraging existing OpenTelemetry paths, Apigee aligns with these evolving standards. This integration ensures that enterprises can scale their agentic ecosystems while maintaining full visibility and control over their distributed API-driven AI workflows.