nCino connects AI agents to its mortgage suite via MCP: banking chooses oversight, not autonomy

By Zendoric · September 28, 2026.

nCino (Nasdaq: NCNO), a provider of software for banking and loan origination, has launched Mortgage MCP, a layer that lets AI agents compatible with the Model Context Protocol (MCP) —an open standard that defines how an AI agent can read and act on the tools of an external piece of software— connect directly to its mortgage suite. According to the company, the goal is for tasks now spread across several screens, simple one by one but slow in aggregate, to be resolved with a natural-language conversation.

The launch includes two preconfigured tools. Admin MCP covers system administration tasks: user role management, state licenses and branch hierarchies. Loan Officer MCP gives origination officers a conversational interface to check loan status, prioritize portfolios, launch income and asset verifications, query the automated underwriting system (AUS) and draft borrower disclosures. Casey Williams, general manager of global mortgage at nCino, sums it up this way: administrative processes that used to take hours become a five-minute conversation, according to the company itself.

What matters is not so much the function as the scaffolding around it. Every action taken by the agent is logged with a timestamp, description and outcome, and the system requires explicit human confirmation before high-impact operations, such as archiving a loan officer or restructuring a branch. The agent always operates within the permissions already assigned to the user invoking it: it has no authority of its own, it inherits the authority that already existed. It is an architecture designed for a sector where any automation touching loan files, officer licenses or customer disclosures immediately draws regulatory scrutiny —in the United States, the Consumer Financial Protection Bureau (CFPB) and a patchwork of state licenses— and where nCino, with more than 2,700 customers among community banks, credit unions and independent mortgage lenders, cannot afford to sell autonomy without traceability.

The use of MCP itself as the connection layer, rather than a proprietary API, is the part that transcends this particular piece of news. The protocol, initially promoted by Anthropic and already adopted by several enterprise software vendors, is consolidating as the standard connector between AI agents and external applications, in a way comparable to how REST APIs standardized the web a decade ago. That nCino builds on MCP instead of a closed channel reduces the friction for a bank already using Microsoft Copilot or Anthropic's Claude to plug them straight into its mortgage suite, with no custom integrations.

This connects with something we have already flagged in our analysis of the dispute among tech giants over control of agent infrastructure: the game is not played only on who has the smartest model, but on who controls the 'plumbing' through which those models get to do real work inside a company. A vertical provider like nCino, which knows the regulatory terrain of mortgages better than any generalist AI lab, becomes a necessary piece of that plumbing: it is not competing to build the best agent, it is competing to be the secure channel through which any agent gets to touch sensitive data.

There is also an employment reading worth keeping in view. The tasks Mortgage MCP automates —license management, portfolio triage, document verification, drafting disclosures— are precisely the kind of administrative and back-office work that, sector after sector, we have identified as the most exposed to AI substitution in banking. The loan officer and the system administrator do not vanish overnight, but the time devoted to the mechanical part of their job does shrink, and with it the need for so much staff dedicated solely to that mechanics. What survives and gains weight is judgment: who decides when to confirm a high-impact operation, who answers to the regulator when something fails, who audits the action log. It is the same rotation we were already seeing in insurance and banking: fewer administrative hands, more risk, compliance and agent governance profiles.

In the long run, this kind of deployment —dull, sector-specific, with no superintelligence headlines— is actually the one that tells us most about where real AI adoption in the economy is heading. The promise of abundance we defend depends on the technology freeing human hours from repetitive tasks so they can go to work that demands judgment and customer relationships; cases like this one, with inherited permissions and every action audited, are the prototype of how that transition can be made without handing control to a system nobody supervises. The question that really matters is not whether the mortgage agent rebels, but who holds the key to authorize it and who reviews the log when something goes wrong.