MCP and API
What Makes a Loan Package Workflow Agent-Ready?
Agent readiness is more than exposing an MCP endpoint. The operator needs discovery, account setup, credentials, a real processing path, results, and credit visibility.
Agent-ready is not the same as MCP-enabled
A product can expose an MCP endpoint and still leave an agent operator stuck. The operator needs to know what the endpoint does, how to create an account, how to create a credential, which tools are available, and what payloads are required for a real package submission.
Loan Intelligence uses public discovery for the human and agent operator, then moves credentials into the authenticated workspace. The public MCP page shows the endpoint, status route, discovery metadata, and agents guide. The workspace key page creates the bearer credential.
That split keeps unauthenticated discovery clear without treating package processing as anonymous infrastructure.
The minimum viable agent workflow
The agent should quote first. A quote tells the operator what the package will cost before any work starts. Processing then reserves credits, stages source payloads, queues the worker, and captures credits only after durable billable outputs exist.
After submission, the agent should poll package status and then retrieve results. Results should include public package state, source references, extracted facts, warnings, tasks, and output links. They should not expose queue names, cache heads, AppData paths, staged payload text, or idempotency internals.
That is the contract behind API and MCP access.
Why source-linked outputs matter
Agent answers are only useful if a human reviewer can verify them. A response that says a lease is missing, a DSCR input is pending, or a policy rule appears outside the box should point back to the source package context.
Loan Intelligence is designed to return source references and generated output links so the agent can summarize without becoming the source of truth.
The human still owns lending judgment. Loan Intelligence provides decision support, not autonomous underwriting or credit approval.
How to evaluate readiness
A real readiness proof should create a disposable account, create an API key, configure an agent-like client, submit a synthetic anonymized loan package, wait for worker and legacy-engine processing, verify billing behavior, retrieve results, and clean up.
A product that cannot prove that flow end-to-end is not truly agent-ready. It may be MCP-enabled, but it is not operationally ready for a lending team or partner platform.
Teams planning an integration can start with Get Started or review the who we serve pages to map the workflow to their operating model.