Enterprise AI Vendor Tray.ai Launches Agent Gateway to Tame Agent Tool Development

Tray.ai has announced the company's Agent Gateway, a new enterprise solution designed to bring order and governance to the increasingly under-managed AI agent tool development space. It specifically addresses a problem companies see as MCP server sprawl and “shadow MCP development”, where a growing number of tools are being built outside IT department supervision.

AI agents increasingly rely on custom scripts and other tools to access non-native data or applications. The integration protocol in question is Model Context Protocol (MCP), and it is by design flexible enough that teams are building their own. This has led to a familiar situation where decentralized innovation is resulting in tool sprawl. Agent Gateway allows MCP tools to be managed centrally and audibly, with consistent governance policies across the organization.

“The pace of AI agent adoption in the enterprise is outstripping existing enterprise policy management processes,” a company spokesperson said. “The features in Agent Gateway provide centralized governance, security and compliance oversight of MCP tools and agents.”

The problem for IT departments is the lack of visibility into this activity. Beyond creating a surface area for security and compliance concerns, these tools and agents have sprawled to the point where IT is now struggling to properly monitor, secure, or provide support for. Agent Gateway addresses these issues by giving teams a central platform to develop, manage, and monitor MCP tools.

The exact mechanisms include policies and guardrails for those with permission to publish tools, what tools they can publish, and what permission levels are required for a tool to be published and used. Policies also govern the versioning and retention of MCP tools in the agent platform.

Agent Gateway gives IT departments visibility into and control over the development and usage of MCP tools in three key scenarios, according to Tray.ai. The first allows the creation of composite MCP tools within Merlin Agent Builder’s graphical user interface. It also allows publishers to leverage over 700 managed connectors in Tray’s own ecosystem to publish connector-backed MCP tools and create MCP tools from Tray-native, prebuilt operations. Finally, it allows teams to securely consume third-party MCP servers while maintaining a single, centrally administered log of all operations that can be audited for insights, errors, and security concerns.

In each of those scenarios, Tray Insights Hub automatically tracks every tool and its execution history to provide context and observability for the entire MCP platform. The system can be plugged into monitoring solutions like Datadog or Splunk to create a consistent data layer for insight, metrics, and log aggregation.

In all, the goal is to give IT departments the visibility and control they need without disrupting workflow. This means granular control over how tools are published, and also which tools are made available to projects and which are locked down. It’s the same problem that any distributed development activity faces in an enterprise context: How do you control who accesses what, and when?

For its part, Tray is positioning Agent Gateway as a natural extension of its current agent management suite, which includes Agent Hub and MAB. That means another layer of centralization, with a managed API that ultimately means that operations teams are monitoring a narrower stack than they would otherwise have in a decentralized environment.

The product was developed to address a very clear market need, however — that of at-scale adoption of AI agents. The change in verb tense is important, because enterprises have been experimenting with AI agents for some time. The fact that we are now seeing pilot programs mature to production use is both the promise and the problem for AI agent vendors.

At-scale use is the point at which governance, security, and compliance considerations are no longer theoretical, and they have to be managed. Agent Gateway is Tray.ai’s bet on what that means in practice: a single platform that can be integrated with a broader set of management and monitoring services, all auditable by operations teams.

The issue for enterprises is whether their AI agent management environment offers enough of a balance between governance and agility to attract IT departments that see decentralized innovation as a major concern. As the playbook from recent years of cloud and API sprawl shows, the model that works is usually some flavor of accepted platforms that provide the security, governance, and policy controls IT teams need without locking out developers from innovation.