Salesforce this week announced “Claudeforce,” a partnership with Anthropic that lets salespeople query customer records and check deals by typing requests to Claude rather than navigating a traditional Salesforce dashboard. The move is part of a broader trend in which large software vendors are rolling out plugins and connectors that allow customers to access corporate data directly through chatbots like Claude and ChatGPT.
What is MCP and why is it spreading?
The Model Context Protocol, or MCP, provides a standardized way for AI models to plug into external systems so that data from a company’s systems can stream into a chatbot. The term has quickly entered corporate discussions and earnings calls. Patrick Starling, who helps lead AI strategy at FactSet, said adoption of the company’s connector has taken off “like a hockey stick” since it launched last year. S&P Global reported that more than 500 customers had connected to its MCP as of last quarter.
Anthropic first introduced MCP in 2024, and AlphaSense data show mentions of the term on corporate earnings calls have risen quarter by quarter globally since then.
Why are companies opening up to AI agents?
Part of the push is defensive: companies fear that if they do not open their platforms to AI agents, AI‑native competitors will build offerings that erode existing business models. Earlier this year investors sold off software stocks on bets that AI agents could undermine traditional models.
Industry participants argue that a firm’s proprietary data — not its website — is the core source of value. As one commentator noted, while anyone can write code to manage schedules or documents, replicating a service like S&P Capital IQ Pro is not feasible in the same way.
Monetization remains unresolved
A central business challenge is how to price and monetize chatbot access to proprietary data. Unlike the prior era of per‑seat billing, there is no industry standard for charging access to data via AI chatbots. A single hedge‑fund analyst can now use agents to dramatically multiply the number of queries they run.
Commercialization is still “evolving,” Starling said regarding FactSet. Peter Licursi, who leads AI strategy at S&P Global, likewise described pricing as “an evolution that we are actively discussing with our customers at all times.” Cristina Pieretti, head of digital content and innovation at Moody’s, said it is still “pretty early,” and Moody’s currently charges based on expected customer usage.
Salesforce did not release pricing details for its new Claude integration.
Token costs and who pays them
Nearly every large company is grappling with rising AI token costs. With MCP‑style chatbot plugins, the end user — for example a business paying for a Salesforce‑Claude integration — bears the additional token expenses associated with the tools. That dynamic complicates matters for financial services providers trying to persuade customers to pay for add‑ons that will increase their token spend.
Some firms provide integrations for free
Not all companies plan to charge for AI integrations. Upwork launched a free MCP server in August that lets users search its job database while remaining in Claude or ChatGPT. As a marketplace that earns only after transactions, Upwork views AI chatbots as “acquisition channels of the future,” CEO Hayden Brown said. Brown added that the integration was also partly defensive after noticing users dispatching AI agents to Upwork that attempted to complete hiring workflows.
Conclusion
By introducing MCP‑based connectors, large software vendors are shifting more work into AI interfaces. Adoption of the technology is growing quickly, but business models — especially pricing and the question of who bears AI token costs — are not yet settled. Over the coming quarters, how companies reshape revenue strategies without diluting the value of their proprietary data will be a key test.
