Uber Builds AI Management Platform 'MCP Gateway'

Uber has independently built a management platform called 'MCP Gateway' to consolidate access points for internal AI agents using various services. The gateway bundles over 800 MCP servers and more than 5,000 AI tools, implementing a unified control mechanism for determining which services agents can access.

The underlying issue is the 'management complexity' emerging as AI agent adoption rapidly expands within enterprises. While AI agents automate tasks across multiple systems on behalf of humans, the growing number of internal services increases the cost of individually managing which agents can use which tools. For large-scale enterprises in particular, tool proliferation and access rights misalignment become serious operational challenges. Even a tech giant like Uber has not been exempt from this problem.

MCP stands for 'Model Context Protocol,' a common standard enabling AI models to integrate with external tools and data sources. Uber's gateway leverages this MCP framework while separating tool registration from actual execution processes. In other words, it decouples the configuration work required when adding new tools from the processing invoked by AI agents in real-time, creating a design that balances extensibility and stability.

Among specific features, API auto-detection stands out first. Internal APIs can be discovered and registered without manual intervention, keeping management overhead manageable even as tool numbers grow. Access control functionality allows narrowing the range of tools available to each agent. Additionally, it incorporates automatic sensitive data masking (automated deletion of sensitive information), designed to mitigate risks of AI agents inadvertently accessing or leaking personal information.

The platform also features what is called 'Omni MCP,' a progressive tool discovery capability. Rather than loading all tools at once, this mechanism gradually narrows the scope as needed, maintaining processing efficiency even when handling thousands of tools. Response filtering functionality is also available, allowing data received by agents to be formatted and restricted according to purpose.

This initiative is positioned as one form of 'governance' necessary for large-scale AI agent operations. While agents offer convenience, insufficient control when integrated with internal systems elevates risks of malfunction and information leaks. Uber's case demonstrates one answer to an industry-wide challenge: how to design systems for systematically managing such risks. As enterprise adoption of AI agents accelerates, future success will increasingly depend on how each company designs and standardizes management layers like gateways.