Google Cloud API Gateway Now Converts REST APIs to MCP Tools
Seamlessly integrate existing APIs with AI agents using Model Context Protocol.
1MCP as the agent standard
The Model Context Protocol (MCP) has emerged as the standard interface for AI agents to discover and invoke tools. Frameworks like Agent Development Kit (ADK) and Gemini Enterprise natively support MCP, while most enterprise capabilities remain locked behind traditional REST APIs. This creates friction as developers currently need to maintain separate MCP servers that duplicate gateway functionality for routing, authentication and quotas.
2Gateway as MCP translator
Google Cloud API Gateway now bridges this gap by acting as a remote MCP server. Developers simply annotate their existing OpenAPI specs with MCP metadata, deploy the configuration, and their REST operations automatically become available as agent-ready tools. This eliminates the need to build, host or maintain separate MCP infrastructure while preserving existing authentication and quota systems.
3Lightweight integration path
Positioned as the lightweight option in Google's gateway portfolio, API Gateway offers rapid integration for services running on Cloud Run. Developers can secure, manage and expose APIs to agents within minutes. For more comprehensive enterprise needs including lifecycle management and monetization, Google recommends using Apigee alongside API Gateway for MCP functionality.
4Transparent request handling
The gateway accepts standard MCP JSON-RPC requests on a dedicated endpoint, transcoding each tools/call into corresponding REST requests. It applies existing policies like JWT or API-key authentication, quotas and logging before translating responses back to MCP format. This maintains a single policy enforcement path regardless of invocation method.
5Implementation workflow
Developers implement MCP support in four steps: annotate OpenAPI specs with tool descriptions that guide LLM usage, deploy the gateway configuration, configure discovery security (defaulting to open access during development), and connect agents to the /mcp endpoint. The gateway automatically handles argument mapping to REST parameters while enforcing existing policies.
6Current limitations
The Public Preview has several constraints: operations returning empty bodies (HTTP 204) aren't exposed, complex nested schemas may render incompletely in discovery, each gateway supports up to 1,000 tools, and MCP can't coexist with model routing in the same config. Google plans to add MCP resources, response streaming and Model Armor inspection in future updates.
7Development example
A sample order service API demonstrates the integration. The getOrderStatus operation becomes an MCP tool named get_order_status, with its description guiding LLMs on when to invoke it. The gateway transparently maps MCP calls to REST requests, maintaining all existing security and monitoring while adding agent accessibility without code changes.
8Strategic implications
This integration significantly lowers the barrier for enterprises to expose existing capabilities to AI agents. By leveraging current API investments and policies, organizations can participate in the growing agent ecosystem without operational overhead. The solution particularly benefits companies with substantial REST API estates looking to adopt AI agent technologies.