The concept of MCP servers is often misunderstood, with many assuming that first-party SaaS MCP servers are the ideal bridges for enterprises to connect to various tools. However, in reality, connecting to per-tool MCP servers is rarely the right design pattern, as it can lead to a fragmented and inefficient system. Instead, enterprises need custom MCP servers that mirror their specific workflow requirements, providing context to AI agents and helping them understand how to use APIs effectively. This approach allows for a more unified and scalable system, where AI agents can work with existing workflows and abide by the rules set by human employees. An internal MCP server can expose tools to AI agents that are mapped to existing workflows, leveraging traditional APIs and eliminating the need for arbitrary paths and security risks. While first-party MCP servers may be useful for consumer applications, they are not suitable for most B2B SaaS software, where tools are often rigged together through human hands, Zapier-like integrations, or code. Ultimately, an internal MCP server is the better approach for enterprises, providing a more practical and theoretically superior solution for leveraging generative AI and getting work done securely.