Uber Builds MCP Gateway to Govern 5,000+ AI Agent Tools
Uber standardizes AI agent integrations using a new gateway for 800+ servers.
1Standardizing AI Agent Connectivity
Uber has developed an MCP Gateway to unify how AI agents discover and invoke internal services. This platform addresses challenges like fragmented integrations and inconsistent security practices as agent adoption scales. By providing a common layer for discovery, authorization, and execution, it supports over 800 MCP servers and 5,000 distinct tools.
2Decoupled Gateway Architecture
The architecture functions as a microservice layer that exposes existing APIs as MCP tools without requiring underlying service changes. It separates the MCP Registry, which acts as a control plane for tool definitions and ownership, from the Proxy Gateway. The proxy handles runtime execution, translating MCP requests into HTTP, gRPC, or TChannel.
3Automated Tool Discovery Process
Uber utilizes an AutoCrawler service powered by Cadence to automate tool discovery. It scans internal interface definition language registries, extracting API definitions from Protocol Buffers and Thrift schemas. A large language model then generates tool descriptions. Newly discovered tools remain disabled until owning teams perform a manual review and explicit enablement.
4Security and Access Governance
The gateway enforces authorization at both server and tool levels using Uber's internal access control system. It includes features for redacting personally identifiable information and sensitive data within tool responses. For third-party integrations like Jira or Google, a separate service manages the secure exchange of internal user tokens for external authentication.
5Optimizing Context for Agents
To handle the growing tool catalog, Uber developed Omni MCP to support incremental discovery. This allows agents to search for servers and retrieve specific schemas only when needed, preventing inefficient context loading. Additionally, Response Projection enables agents to request specific fields rather than receiving complete payloads, further optimizing performance for complex tasks.
6Industry Parallels in Gateway Design
Industry experts note that Uber's approach mirrors established API gateway patterns used for decades. Microsoft's Ajay Sharma and Pipe17's Kelly Goetsch compared the architecture to existing solutions for APIs and GraphQL. These systems provide necessary order to the inherent chaos of LLM integrations, emphasizing the importance of established distributed systems governance.
7Developer Tooling and CLI
Uber introduced the aifx command-line interface to assist coding agents in discovering and invoking tools. This tool allows developers to write execution results directly to files for selective inspection. Configuration changes propagate to the data plane dynamically, ensuring that updates do not require service restarts or redeployments, maintaining high system availability.
8Strategic Shift in Infrastructure
This initiative highlights the growing necessity for robust governance in AI-driven development environments. By treating AI agents as first-class citizens within a microservice architecture, Uber ensures scalability and security. The focus on observability and standardized discovery marks a significant evolution in how large-scale organizations manage complex, agentic tool ecosystems today.
Original article: InfoQ — Uber Builds MCP Gateway to Govern 5,000+ AI Agent Tools