.svg)
.webp)
While AI ambitions are high, trust in the technology is still in flux. Many companies can’t confidently let AI agents automate routine tasks, like resetting a password, approving an order, or running a transaction. Instead, they struggle with getting agents to accurately and securely perform their duties the same way, every time.
This is common in technology transformations of this magnitude. For example, new protocols and governance had to be developed before companies could harness the power of the Internet in their own operations. Similarly, before organizations can successfully deploy and get value from AI agents, they must set up guardrails that enable the systems to have broad, secure access across the organization’s IT environment.
Several recent protocols aim to deliver this orchestration, but Model Context Protocol, or MCP, is emerging as the top choice. More than just IT plumbing, MCP increasingly serves as a glue connecting AI agents to core enterprise systems to transform common, multi-step workflows, like order-to-cash or record-to-report. Released by Anthropic in late 2024, there are already thousands of open-source MCP servers to choose from — reflecting the important role MCPs play in enterprise AI adoption. But as businesses rush to quickly infuse the power of AI into their operations, they must be careful about the foundation they choose.
"At Workato, we hear every day that while MCP is exciting, enterprises still face challenges making MCP work securely, effectively, and reliably at scale," Adam Seligman, Chief Technology Officer at Workato, said in Axios.
Many of the available MCP servers require companies to build necessary components themselves. Key requirements, like security, identity and access controls, and governance are custom add-ons, not features. Often, engineers start from scratch to rebuild environments every time. This complicates the IT environment. The AI agent is stuck navigating hundreds of API calls, leading to unreliable answers and actions.
