MCP isn’t KYC-ready: Why regulated sectors are wary of open agent exchanges

Since its debut in November, the Model Context Protocol (MCP) has quickly attracted a significant user base, raising expectations for its potential to become an industry standard. However, certain sectors, particularly regulated industries like banking, are proceeding with caution. While many financial institutions have been at the forefront of AI innovation, employing advanced machine learning algorithms, their immediate adoption of MCP and Agent2Agent (A2A) technology is far from guaranteed. Many banks and financial service providers have started to experiment with AI agents, but these initiatives primarily involve internal systems. While APIs are available, the integration processes that regulated entities undertake are lengthy and require rigorous compliance checks. "It’s very early days in a rapidly evolving field, but there are fundamental building blocks still missing, especially regarding interoperability and effective communication," explained Sean Neville, co-founder of Catena Labs. Drawing parallels to the early internet, Neville noted that the lack of secure transaction protocols like HTTPS hindered the creation of e-commerce platforms until those fundamental technologies were established. Today, as enterprises and AI platform providers develop MCP servers for multi-agent systems, the importance of foundational standards is more evident than ever. MCP allows for agent identification, which helps servers ascertain the tools and data at their disposal. However, many financial institutions remain cautious, seeking greater assurance that they can control integrations and restrict the sharing of sensitive information. John Waldron, Senior Vice President at Elavon, a subsidiary of U.S. Bank, mentioned in an interview with Venture Beat that while they are exploring MCP, many uncertainties linger regarding the protocol’s standards. "There aren’t many standardized solutions emerging, so we are exploring various methods, possibly even bypassing MCP if the agent technology is compatible between different domains," Waldron said. He emphasized the need for traceability in data exchanges to avoid risks associated with data leakage. Regulated businesses are not unfamiliar with AI, as the rise of robo-advisors has demonstrated. These algorithms enable significant advancements in passive investment strategies. Moreover, banks have invested in natural language processing for improved document analysis. However, integrating new AI models with existing risk management frameworks has proven challenging. Greg Jacobi, Vice President at Salesforce, highlighted that many financial clients have established processes for model assessment but face difficulties when introducing LLMs (Large Language Models) into their systems, which often yield unpredictable results. Jacobi remarked, "These firms typically expect consistent outputs from their models; any deviations raise red flags, necessitating stringent quality control measures." Although regulated companies have adopted APIs, they often hesitate to embrace public-facing technologies that could compromise their control. Nevertheless, Waldron remains optimistic about the future of MCP and A2A within financial institutions, noting its potential significance in evolving business logic. As Catena Labs’ Neville observes the ongoing discussions around interoperability protocols, he believes AI agents will eventually become as valuable to banks as human customers. Given his background in launching the USDC stablecoin, he understands the hurdles of introducing new technologies into regulated environments. While MCP is open-source and continuously evolving, it still lacks essential features like communication guardrails and audit trails. These gaps could be addressed through MCP, A2A, or entirely different standards such as LOKA. One of the significant challenges with MCP lies in authentication. When AI agents enter the financial ecosystem, the lack of a reliable method for "know-your-customer" (KYC) checks raises concerns. Financial entities require assurance that these agents engage with licensed counterparts, necessitating a verifiable identity framework. Neville emphasized that establishing a method for agents to assert their identity and associated risks will be crucial for future developments in agentic frameworks.