For most of the past few decades, security operators had to apply the same basic logic to all functions of enterprise security software: They had to seek out the right module, enter the required data and prompt the system to execute a predefined action. The logic applied to most tasks—from pulling video footage during forensic investigation to entering a new user profile and supervising access control logs, but increasingly also booking a meeting room, helping organizations optimize space usage and other functions beyond the traditional realm of security.
In all cases, the operator had to know the structure of the system before he could access its functions.
With the upcoming integration of ZKBio CVSecurity, ZKTeco is ushering in a new operational logic. By integrating the Model Context Protocol (MCP) and AI Agent capabilities, ZKTeco’s flagship unified security and smart-space platform is moving from a system that must be operated to a system that can simply be told what to do.
Think of a routine request: a client is expected to visit the premises of the company, and a meeting room, visitor access and a parking space need to be arranged. Previously, the operator had to open several software modules in a predefined sequence, in many cases requiring that they enter the same information multiple times.
With ZKBio CVSecurity, the full sequence of overlapping actions can be folded into one single prompt. The system determines autonomously who is arriving, when and where the meeting can take place, and whether visitor credentials need to be issued and a parking space be reserved.
“The future of enterprise software is not necessarily more functions,” explains Mrs. Popy Xiao, Senior Product Manager, ZKTeco. “It is software that understands user intent more accurately.”
Building a platform for AI agents and human users
ZKBio CVSecurity already brings multiple functions of security—as well as functions beyond the traditional realm of security—under a single operational view: From people and vehicle access at the company gates, to visitor management, time and attendance, parking, video surveillance, intrusion alarm and meeting-room management. So far, these capabilities have been optimized for human operators using a web console or mobile app.
With this evolution, ZKTeco is adding a second mode of use, enabling AI agents to understand those same capabilities and act on them autonomously. MarsLume AI works as the intelligence layer within ZKBio CVSecurity—handling natural-language understanding and multimodal interpretation behind the scenes—while ZKBio CVSecurity provides the business execution layer. The platform thus takes on a dual role: it remains the operational system security teams use through learned routines and intuitive workflows, while simultaneously becoming the execution layer that lets AI agents automate processes in true partnership with the human user. All user interactions and operations stay within a single, unified platform.
MCP: the bridge between AI and business execution
With the advent of large-language models, AI has become capable of understanding natural-language requests and acting on them accurately. However, AI models usually lack the insight into live operational data of the organization that is deploying them. They lack the authority to open door, create visitor credentials or reserve a meeting room. For example, AI models can parse questions as to whether a meeting room is free at 3pm, but only through an autonomous, agentic connection to the underlying system can an AI model book a room.
This is the gap MCP is designed to close—not on a piecemeal, but on a system-integration level. Instead of building a dedicated integration every time a new AI application needs to reach an enterprise system, MCP exposes ZKBio CVSecurity's capabilities in a standardized form that an AI can understand and invoke directly. This approach not only allows organizations to govern AI agent access efficiently, but it also enables AI agents to safely execute tasks such as querying personnel and visitor records, creating credentials, checking and reserving rooms, looking up vehicles and parking records, or retrieving video events—without an integrator having had to set up the system for each specific request in advance.
In simple terms, MCP works like a USB interface for the AI era: it gives AI a standardized way to discover and use the capabilities of ZKBio CVSecurity.
Importantly, every MCP call is governed by the same role-based access control (RBAC) that applies to human users. An AI agent cannot perform actions beyond the permissions granted to its assigned role, and every operation—whether initiated by a person or an agent—is logged with full audit trail. This means organizations do not need to build a separate governance framework for AI; the security policies already in place extend to agent-driven workflows automatically.
From Tools to Skills: organizing capability around real workflows
Creating a unified connection framework for AI is not enough on its own. If each capability is unlocked individually, without the greater operational context, AI models would have to work out, for every request, which combination of Tools they should apply. ZKBio CVSecurity addresses this with an intermediate layer ZKTeco calls Skills: task-oriented groupings of Tools aligned to the workflows of operators—regardless of whether they are human or AI—rather than the architecture of the underlying software.
A Visitor Management Skill, for instance, connects visitor credential creation, visit-window configuration, access-rights assignment and parking arrangement into a single coherent capability. A Meeting Management Skill covers room availability, conflict checking, reservations and the access privileges that go with a booking. A Security Investigation Skill can combine relevant alarms, video events, video retrieval and people-and-vehicle association to help operators investigate an event.
Users do not need to know that these underlying tools exist individually—they only need to describe the outcome they want.
From Skills to Agents, from capabilities to goals
ZKTeco draws a clear line between Skills and Agents. A Skill answers what the system is capable of, while an Agent answers what needs to be done to complete a goal. The latter might involve multiple Skills, as is typical in real-life scenarios.
Take, once more, the example of a client visit: A Visitor Management Skill alone does not know how to check room availability, and a Meeting Management Skill alone does not know how to issue visitor credentials. An AI Agent orchestrating both Skills breaks the sequence down into its component tasks: identify the visitor, check the proposed time, find and reserve an available room, create the visitor record, configure door access and arrange parking if needed. The result is one consolidated set of actions covering all that is needed to facilitate the visit.
From the user's side, one sentence goes in; from the system's side, several Skills and capabilities are being invoked to answer it.
“The user sees one sentence,” Mrs. Xiao states. “Behind that sentence, multiple Skills and multiple business capabilities may already have been invoked.”
Case in point: from video retrieval to event investigation
The integration of video security illustrates the shift particularly well. Conventional workflows assume the operator already has a rough idea of when and where an incident occurred—enabling him to narrow down where within the large body of footage the needed evidence can be found. With an AI Agent layered on top of ZKBio CVSecurity, operators can instead ask for any footage of “unusual activity near the perimeter lately,” for example. The Agent interprets the broad terms used to describe time and place, as well as what might constitute “unusual,” and then invokes all the required event queries, video retrievals, target recognition operations and people-and-vehicle associations on the user's behalf. What the Agent present to the user is not hours of footage to review, but an event the system has already made sense of.
The same approach can also support alarm handling by bringing relevant video context and business information into the operator's workflow, with further agentic automation as capabilities evolve.
Final thoughts: Complex architecture, simple experience
New workflows enabled by ZKBio CVSecurity do not intend to replace human operators. Instead, the focus is on lowering the barrier to operating inherently complex systems: A large smart space can involve dozens of subsystems and a dense catalogue of functions, and ordinary operators should not need to know which system owns which task or how many steps a job requires. Behind the scenes, that means AI models, MCP, Skills, Agents and ZKBio CVSecurity's underlying business systems are all doing their part. What the end user sees is meant to stay simple: a single sentence describing what needs to happen, with the platform handling the rest.
Or, in the words of Mrs. Xiao: “The industry does not lack software modules. What it has lacked is an operating model in which people can state an operational goal and the platform can complete the corresponding work across systems, under policy and under audit. That is the product direction we are taking with ZKBio CVSecurity, MarsLume AI, MCP, Skills and Agents. The architecture is sophisticated by necessity. The experience offered to the customer should remain simple.”
For more information about ZKTeco products and solutions, please visit
www.zkteco.com.