Model Context Protocol (MCP) has quickly become a baseline expectation in enterprise software, with the ecosystem growing to approximately 20,000 public servers within 18 months. Customers now expect to query data, run analysis and initiate actions on vendor platforms through agentic AI harnesses such as Claude, ChatGPT and Copilot, and MCP has moved onto the product roadmaps of most SaaS companies as a result.

For vendors, however, MCP is commercially challenging. Servers carry a cost per API call and are expensive to build and maintain, with some estimates suggesting a single enterprise read-only server can cost as much as $300,000 to build, while customers tend to expect MCP access to be free and often underestimate how its cost rises at enterprise scale. As AI harnesses become the primary interface for enterprise software, vendors also face the risk of perceived value shifting away from their platforms.