Uber has deployed an MCP gateway that standardizes how AI agents discover and invoke internal services across more than 800 Model Context Protocol servers and 5,000 tools1,2. The platform addresses fragmented integrations, duplicated infrastructure, and inconsistent security practices as agent adoption expands across the company.
Control plane and runtime split
The architecture exposes existing APIs as MCP tools without requiring changes to underlying services. A microservice design separates the MCP Registry, the control plane, from the Proxy Gateway, the runtime execution layer. The MCP Registry maintains tool definitions, schemas, ownership records, and enablement settings. The Proxy Gateway translates MCP requests into HTTP, gRPC, or TChannel calls and converts responses back into MCP-compatible results.
AutoCrawler automates tool discovery using Cadence, Uber's workflow engine. It scans internal interface-definition-language registries and extracts API definitions from Protocol Buffers and Thrift schemas. A large language model produces tool descriptions from those schemas. Newly discovered tools remain disabled until their owning teams review and explicitly enable them.
Security and governance layers
The gateway enforces authorization at both server and tool levels through Uber's internal access-control system. It supports redaction of personally identifiable information and other sensitive data in tool responses. For third-party integrations including Jira and Google, a separate service exchanges internal user tokens for corresponding third-party authentication tokens. Configuration changes propagate to the data plane without requiring service restarts or redeployments.
Uber also developed Omni MCP to support incremental discovery: agents can search for servers, discover relevant tools, retrieve individual schemas, and invoke selected tools as needed. A command-line interface called aifx lets coding agents discover and invoke tools and write results to files for selective inspection. Response Projection lets agents request specific response fields instead of receiving complete payloads.
Industry comparisons
The architecture has drawn comparisons with established API gateway patterns. Ajay Sharma, Principal Software Engineer at Microsoft CoreAI, wrote in a LinkedIn post that "Azure API Gateways has been doing this for a while now". Kelly Goetsch said Uber's gateway resembles gateways built for APIs and later GraphQL. Tamir Dresher said security, discovery, observability, and governance remain central concerns regardless of protocol layer.
ANALYSIS The gateway pattern treats MCP as an enterprise integration layer rather than a novel protocol, routing agent requests through the same proxy-and-registry architecture that large organizations already use for API management. The scale — 800-plus servers and 5,000 tools — makes it one of the largest disclosed MCP deployments to date.