Google has expanded its Model Context Protocol (MCP) support to Apigee hybrid, enabling customers to deploy agentic tools within their own Kubernetes clusters. Announced on October 14, 2026, this update allows Apigee hybrid version 1.17.0 to run the MCP data plane inside a user’s own infrastructure, addressing a key demand from customers who prefer to keep their runtime in private data centers or other public clouds due to regulatory or operational requirements.
The MCP support in Apigee hybrid maintains the same conceptual model as the original Apigee version. Users describe their APIs with an OpenAPI 3.0.x specification, create an MCP Discovery Proxy, and make every operation available as an MCP tool. Agents interact with the endpoint via JSON-RPC, discovering tools with tools/list and invoking them with tools/call. The key difference in the hybrid version is that the MCP data plane operates entirely within the user’s Kubernetes cluster, ensuring that request traffic never leaves the cluster and configuration is pulled rather than pushed.
This architecture ensures data residency and sovereignty, as all MCP request traffic stays within the user’s infrastructure. The governance and security policies already in place for Apigee APIs apply to agentic endpoints, including OAuth 2.0, OpenID Connect, and API key verification. Additionally, the MCP Discovery Proxies are automatically ingested into Apigee API hub, making them searchable and integrable with existing API performance monitoring tools.
To enable MCP in Apigee hybrid, users must set the enableMcpServer flag to true in their overrides.yaml file and upgrade the apigee-operator and apigee-org charts. Google emphasizes that this should be done during an approved maintenance window, as enabling or disabling MCP triggers a staged canary release of runtime pods. The company also highlights future plans to improve end-to-end tracing of agent calls across AI clients, Apigee, and backend systems.