Netskope adds MCP controls to secure enterprise AI use
Netskope has introduced new security features that monitor and control Model Context Protocol (MCP) traffic inside its Netskope One platform, as enterprises expand the use of AI agents linked to internal systems.
The update targets organisations that connect AI agents to corporate data and tools using MCP. The protocol is gaining traction as a standard for agent-based AI workflows in businesses.
MCP allows AI agents to request data, trigger actions, and coordinate tasks across different applications. The same features also create new security exposures when agents interact with sensitive data or trigger automated actions.
Netskope said the new features provide visibility into MCP communications and enforce access policies on them. The company said this covers both MCP servers and MCP clients running within an organisation.
The platform can now identify MCP servers and clients in use. It lists attributes such as name, ID, URL, version, host, data source, and protocol.
Netskope is also extending its Cloud Confidence Index (CCI) risk scoring system to MCP servers. CCI is a scoring method that Netskope uses for cloud and SaaS services.
The expanded CCI scoring gives a risk rating to MCP servers that AI agents connect to inside an organisation. Netskope said this aims to help security teams prioritise which AI tools, agents, or integrations pose the greatest risk.
Security teams can then use policy controls inside Netskope One against those MCP endpoints. They can allow or block MCP traffic based on the assessed risk and context.
The company said the platform supports a default block option for MCP traffic. Administrators can then grant access in specific, approved cases.
Netskope said its controls follow a least-privilege approach. That means it can restrict AI agents and tools to the minimum access needed for their tasks.
The features also detect and monitor non-human traffic between MCP servers, clients, tools, hosts, data sources, and development tools. The platform treats this as a distinct traffic category inside its monitoring interface.
Netskope said it can log MCP events in detail. The logging covers sessions, initialisations, tool requests, responses, and deployments.
The system also inspects data that flows through MCP tools. It can identify sensitive information such as intellectual property and passwords within that traffic.
This inspection allows data loss prevention rules on MCP traffic. These rules can block or quarantine transmissions that contain certain types of content.
The company said these functions sit alongside existing Netskope One capabilities. These include real-time traffic inspection, access control, and data protection across cloud, web, and private applications.
John Martin, Chief Product Officer at Netskope, said AI adoption is now a core focus for most enterprise teams. He said MCP now sits at the centre of many of those efforts.
"Every team wants to confidently accelerate AI adoption, and emerging protocols such as MCP are now fundamental to that discussion," said John Martin, Chief Product Officer, Netskope. "MCP also creates new security risks that legacy tools can't solve. That's why we're further extending the capabilities of Netskope One to enable teams to see and create policies for MCP traffic and immediately assess how risky MCP tools are. This is critical to the secure use of AI as organisations develop agents to drive business productivity."
MCP has gained visibility as vendors and developers create servers that expose databases, SaaS platforms, and internal tools to AI agents. Many of these servers are publicly available and can connect to a wide range of models and agent frameworks.
Security leaders have expressed concern that unmanaged use of MCP could expose sensitive corporate assets. They have also warned that autonomous commands from AI systems could affect production systems if they are not constrained.
Vendors are responding with products that treat AI agent traffic as its own category. They are building controls that resemble those for cloud and SaaS use, but applied to AI interactions.
Netskope said the new MCP security capabilities are now in preview for its customers. The company expects general availability in the first half of 2026.
