Uber Builds an MCP Gateway for Internal AI Agents
Uber has built an MCP Gateway to standardize how internal AI agents discover and invoke services, according to InfoQ 中文 AI&大模型. Uber says the system supports more than 800 MCP servers and 5,000 tools through a Registry and Proxy Gateway, with protocol conversion, authorization, schema generation, and progressive discovery; these scale figures remain vendor claims reported by one outlet.
The disclosed architecture offers a production-scale reference for governing agent access to large internal tool catalogs, including authorization, review, data masking, and selective tool discovery.

随着 AI 智能体在公司内部得到广泛采用,Uber 构建了一套 MCP Gateway,用于规范智能体发现和调用内部服务的方式",以解决集成碎片化、基础设施重复建设和安全实践不一致等问题。该平台支持 800 多个 MCP 服务器和 5000 多个工具,为服务发现、协议转换、授权和执行提供统一层。
Uber 工程师将该方案描述为一种微服务架构,可以将现有 API 公开为 MCP 工具,而无需修改底层服务。整个架构将作为控制平面的 MCP Registry 与负责运行时执行的 Proxy Gateway 分离。Registry 维护工具定义、Schema、所有权和启用设置;Proxy 则将 MCP 请求转换为 HTTP、gRPC 或 TChannel 调用,再将响应转换回兼容 MCP 的结果。

#Uber#Uber MCP Gateway#Uber MCP Registry#Uber Proxy Gateway#Uber AutoCrawler#Uber Omni MCP#Uber aifx#Cadence#Microsoft CoreAI#Pipe17