Blend Labs has announced the launch of Autopilot MCP, a server built on the Model Context Protocol, an emerging open standard for AI agent connectivity, that gives authorized agents secure, programmatic access to the Blend platform.

For lenders and partners, Autopilot MCP introduces a new category of capability: the ability to build and deploy AI agents tailored to their workflows, guidelines, and borrower experiences without rebuilding underlying infrastructure.

Solving the orchestration problem in lending

Before MCP, deploying AI in lending required a separate integration for every system an agent needed to touch. And in mortgage, that’s dozens of systems before a single loan closes: credit bureaus, pricing engines, underwriting platforms, title companies, compliance tools, disclosure systems, most of them built decades apart and never designed to work together. Each new connection brought its own engineering project, its own security review, its own compliance sign-off.

With Autopilot MCP, any agent, Blend-built, lender-built, or partner-built, can access Blend’s origination stack through a single interface, from credit and underwriting through compliance, disclosures, and closing.

The release introduces several core capabilities:

• Agentic workflow execution: Rather than surfacing information that requires manual follow-through, agents can execute lending workflows — pulling credit, checking pricing, and verifying compliance, preparing a sequenced submission for the loan officer to decision.
• Institution-specific configuration: Each agent operates against that lender’s own data, guidelines, and loan workflows inside Blend. Lenders running portfolio products, HELOCs, or proprietary overlays can apply institution-specific rules in place of, or alongside, standard GSE guidelines.
• Continuous platform updates: Because Autopilot MCP operates through a standardized interface, new Blend capabilities are available automatically to every lender with Autopilot activated. This means no upgrade cycles or implementation projects.
• Built in access controls: Every agent action is logged with a full audit trail. Access is controlled at the lender level, credentials are isolated per deployment, and shuts down access if a control layer is unreachable. Destructive operations (rate locks, credit pulls, disclosure delivery) are gated until a lender is ready to enable them.

“Until now, the hardest problem in lending AI wasn’t the intelligence of the models,” said Nima Ghamsari, Head of Blend. “It was getting them connected to the right systems, with the right controls, in a way a bank’s compliance team could actually approve. Autopilot MCP solves that. The intelligence is customizable, the infrastructure is shared. Lenders aren’t buying a specific AI feature. They’re getting a surface they can program.”

Autopilot MCP is built on Model Context Protocol, introduced by Anthropic in 2024 and now the open standard for AI agent connectivity across the enterprise software industry. With this release, Blend brings that standard to lending.