AI
Image: Treecha/Shutterstock

Vegangster is pushing the Model Context Protocol (MCP) as a standardised way to connect artificial intelligence agents to platform functionality, allowing the platform to be open for partners to customise it their own way.

As AI becomes more integrated with iGaming operations, Vegangster Chief Product Officer Michael Oziransky details how MCP works, what platforms need to support reliable agent interactions and why the technology could eventually contribute to a shift towards headless iGaming platforms.

Describing MCP

Defining MCP, Oziransky stated: “MCP is just a way for the AI agent to access the platform. It’s a protocol that provides context, basically an explanation of what it means for the agent to call a certain tool. A tool is like an API call, so the protocol standardises access for agents.”

However, he emphasised that once agents begin interacting with platform systems, access must be managed responsibly, with AI operating under the same permission framework of proper traceability of their actions, the same as regular users.

Oziransky added: “It doesn’t matter if an agent was sent on behalf of a certain user. At the end of the day, it’s just a tool that helps that specific user. Nothing should happen that wasn’t expected, especially in terms of permissions, governance and access to sensitive data.”

MCP must be connected to the platform’s existing permission management instead of as a separate entity with its own set of access rules.

“You need to reuse the existing set of permissions. That way, you won’t have a situation where you develop a certain feature, enable access and forget to close certain things because it’s a different path for accessing the platform,” said Oziransky.

He also explains that, as “a matter of coverage”, MCP should allow agents to perform at least the same tasks as a regular user through the platform’s interface. 

Understanding and order

Oziransky stated that another important part of an effective MCP implementation is the instructions given to agents, as while they can read technical descriptions of MCP tools, they also need additional context to understand how to use them and in what order to perform particular operations.

To avoid agents having to explore the available tools and work out which ones to use and in what order, Vegangster instead works on explaining the sequences involved in specific procedures.

“The agent starts to guess, and then it’s not efficient in terms of tokens and token economy. It’s time-consuming and creates a lot of stress on the platform because you’re recalling certain tools without any specific logic,” Oziransky explained.

This is why there is more of a focus on documenting procedures and how the available tools should be used, instead of technical documentation alone.

Traceability, permissions and compliance

Another central requirement for operators using MCP is monitoring agent activity. Oziransky noted that an AI agent works like an employee who wants to complete the assigned task but may keep trying when something does not work as expected.

“It can try endless times to do a certain thing. It produces a lot of access requests and actions, and you need to make sure that everything is 100% traceable,” noted Oziransky.

Three main areas were highlighted that need to be addressed by operators: traceability, permissions and thresholds. This includes recording agent actions, monitoring their behaviour and setting alerts that can close a connection if an agent exceeds established limits or performs too many operations.

Data governance must also reflect the regulations applicable to each market, with European requirements concerning the transfer of certain data to models located outside the European Union given as an example.

Oziransky stated: “In some cases, you need to make sure that certain data isn’t sent to models located outside the European Union. There will also be cases where data isn’t fully open, but masked or anonymised.”

He added that operators should understand which models are being used, where they are located and whether they are public or private, considerations which must be addressed in accordance with the specific regulations under which a business operates.

Headless iGaming platforms

Looking ahead, Oziransky believes iGaming platforms could eventually move towards a headless model, in which the underlying technology and functionality are separated from the interface used to access them, unlike in traditional software architectures.

Operators would be able to connect to the platform’s core through different agents or APIs, rather than relying on a single back-office system, allowing them to have greater flexibility to build tools while relying on the same infrastructure for platform operations, security and data management.

Oziransky commented: “I imagine a totally headless platform. You have a very powerful core with a lot of functionality, which is robust and has all the features we’ve discussed, especially around security, data consistency and permissions.”

He points to a similar trend among enterprise software providers and sees no reason why iGaming platforms would not move in the same direction. However, he expects the transition to require time and education.

For now, Vegangster is focusing on supporting both traditional back-office interfaces and agent-based access. Oziransky believes some operators already use their own agents and solutions to interact with the company’s tools without using the back-office interface, while others continue to configure their platforms through the traditional UI.

Ultimately, he thinks the technology should be introduced to address operational needs rather than simply to associate a product with AI.

Oziransky concluded: “The end goal is for all operators to build a successful business. The tools aren’t there just for the sake of putting an AI label on the banner. They need to do the work they’re supposed to do more efficiently than it could be done through human effort.”