JFrog MCP Registry: Governing AI Agents at the Tool Level
Enterprise Developer PlatformsJFrog’s MCP Registry solves a problem that every platform team running AI agents will hit: security review is a bottleneck, but skipping it means trusting whatever an MCP server can do to your production environment.
Context
Model Context Protocol has become the de facto standard for connecting AI agents to tools, data, and services. Google Cloud teams building on Gemini 3 and the Agent Development Kit have access to a rapidly growing ecosystem of MCP servers — but each one introduces a new trust boundary. Without governance, the choice is binary: block everything and slow agent development, or approve blindly and accept unknown risk. JFrog’s MCP Registry is designed to break that stalemate.
The Detail
JFrog MCP Registry applies the same governance model that Artifactory applies to software binaries — now extended to MCP server deployment. The core architectural decision is enforcement at the individual tool level rather than at the server level. Platform teams can approve specific MCP capabilities while blocking others on the same server, which prevents the over-privilege problem that emerges when server-wide approval grants agents more access than their task requires.
JFrog Curation handles the automated vetting: every MCP server is scanned against security, compliance, and operational policies before it can enter the registry. Non-compliant or malicious servers are blocked before developers can pull them. For engineers, the self-serve workflow runs through a CLI gateway that integrates directly with Cursor, VS Code, and Claude Code — providing access to pre-approved local and remote MCP servers without a ticket or a waiting period.
The Google Cloud integration connects this governance layer to Gemini 3 agents working with BigQuery, Google Kubernetes Engine, and Cloud Storage within Google’s Agent Development Kit (ADK) ecosystem. The registry provides the trusted intermediary between ADK’s orchestration layer and the MCP servers those agents consume. JFrog frames the problem it solves in three categories: supply chain exposure from malicious MCP server code, over-privileged agent access from server-level permissions that grant too much, and shadow AI from unsanctioned MCP servers used outside any governance boundary. A single control plane addresses all three.
The Pattern
JFrog has published four posts in the past 90 days on AI software supply chain governance — covering trust and traceability in AI model supply chains, governed AI model supply chain on AWS SageMaker, compliant version selection via Curation, and MCP tools for AI agents. The recurring angle: MCP servers are the new packages, and the governance discipline that Artifactory applied to binary artifacts must now apply to AI agent capabilities and tools.
The So What
Platform teams already standardized on JFrog for artifact management now have a path to extend that governance to AI agent tooling without adopting a separate security product. The tool-level permission model is the key differentiator from simpler allow/block approaches: it lets organizations deploy agents with narrow, auditable access rather than broad server permissions that drift toward over-privilege as agent capabilities expand. For teams building on Google Cloud’s ADK, the registry provides the missing governance layer between Gemini 3 orchestration and the MCP ecosystem.
Photo by Tima Miroshnichenko on Pexels
Source: JFrog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →