Uber built MCP Gateway to scale its use of MCP for AI agents. It centralizes MCP logic into a single gateway for all agent-to-service interactions. The platform includes a control plane with an MCP Registry and AutoCrawler for discovery. A data plane handles runtime execution and protocol translation between MCP, HTTP, gRPC, and TChannel. MCP Gateway currently hosts over 800 MCP servers and 5000 tools. Service owners must explicitly enable each MCP server and tool after review. The gateway enforces authorization, rate limiting, and sensitive data redaction. Third-party MCP servers like Jira and Google are also supported. The system dynamically updates configurations without restarts.
What changed
Teams built MCP integrations independently, leading to fragmented tooling and duplicated infrastructure.
What it unlocks
Centralized, scalable MCP management for AI agents to interact with Uber’s back-end systems uniformly.
- 800+ MCP servers hosted
- 5000+ MCP tools available
- thousands of internal services integrated
What you need to act on it
- Uber’s internal service ecosystem
- MCP-compatible agent workflows
Sources