SurePath AI unveils real-time controls to secure MCP use

SurePath AI has introduced Model Context Protocol (MCP) policy controls that apply real-time rules to the MCP servers and tools used by generative AI clients and agents.

The release is positioned as a security and governance response to MCP's growing use in workplace AI workflows. The controls determine which MCP endpoints and tool functions can be accessed-and what is blocked-before an AI-driven request is executed.

MCP has become a common way for AI applications to connect to business systems. Desktop and developer tools such as ChatGPT, Claude and Cursor can use MCP to reach local and remote services. These connections can include internal repositories and administrative interfaces, such as Google Drive, Salesforce and AWS management APIs.

This connectivity changes the risk profile for organisations adopting generative AI. AI clients can take actions that run under an end user's identity and privileges. Local MCP tools can run on a user's laptop and, in some desktop applications, may be started without prominent prompts. Remote MCP servers add exposure by increasing the number of network paths and endpoints involved.

Policy controls

SurePath AI's MCP Policy Controls govern which MCP servers and tools are permitted across an organisation. They apply checks in real time, before a tool is made available for an AI workflow to call.

Randy Birdsall, CPO and Co-Founder at SurePath AI, said MCP adoption is following earlier waves of generative AI uptake inside organisations.

"MCP has quickly evolved from a buzz-acronym to the backbone in next-gen AI-powered workflows," said Randy Birdsall, CPO and Co-Founder, SurePath AI.

"In fact, we are seeing the same pattern when ChatGPT first became available - rapid adoption, little oversight, and a surface-level understanding of risks. The reality is that MCP introduces an entirely new attack surface, one that many organizations are already exposing without realizing it, but blocking MCP is not practical. Instead, it needs to be managed securely, and that means moving beyond traditional firewall and IAM policies. Modern organizations need to put into place controls that are specific to how MCP operates. Only then can security teams confidently support AI adoption without hindering innovation."

SurePath AI described MCP as a direct connection between generative AI clients and operational business systems. It said a mix of local and cloud-based MCP servers can create complex connectivity patterns, increasing the risk of data sprawl and lateral movement in the event of compromise.

Discovery and enforcement

The approach combines discovery, classification and enforcement. It monitors MCP usage across AI tools, intercepts MCP payloads, and applies allow and block policies to the tools referenced in those payloads.

One element is MCP tool discovery. Tools that are blocked by policy, or fail capability requirements such as being read-only, are removed from the payload before it is forwarded. As a result, the backend service does not receive a request that includes access to a disallowed tool.

The controls include a block list for tools discovered in an environment and an allow list for permitted tools. An "Allow Read-Only" setting can automatically enable read-only tools without requiring manual addition to the allow list, according to SurePath AI.

Another option is a "Catch-All Action" setting, which determines what happens when a tool is neither explicitly allowed nor blocked. SurePath AI said this provides control over how the system handles tools outside the defined lists.

Auto-discovery and classification are also included. SurePath AI said it can indicate whether an MCP tool is well known or built locally, which can affect governance and risk assessment.

Remote catalogue

On the remote side, SurePath AI said it maintains a catalogue of known MCP servers and endpoints. Protected MCP traffic is routed through its platform for real-time access control, and policies can be applied at the level of an individual tool.

SurePath AI also said the capability can detect previously unseen MCP tools. It framed this as a way to identify supply chain threats, including tools that impersonate others or attempt to move data outside an approved perimeter.

The announcement comes as security teams shift focus from model usage to the actions AI systems can take through connectors and agent tooling. MCP adoption has grown alongside wider use of AI assistants for coding, document handling and workflow automation, which often require access to internal systems and data stores.

SurePath AI said the policy controls are enforced before execution, allowing organisations to set specific rules for MCP servers and tool usage as MCP expands in day-to-day AI workflows.