Product Designer

The Mortgage Office and DossDocs announce new integration partnership to expand document generation capabilities for lenders. The Mortgage Office August 5, 2026 Huntington Beach, California, August 5th, 2026 - The Mortgage Office (TMO), a leading provider of loan origination and servicing software for lenders, today announced a new integration with DossDocs, a comprehensive on-demand loan document platform built by experienced mortgage lawyer Dennis H. Doss. The partnership expands document capabilities available to TMO customers, providing lenders with additional flexibility on how they generate loan documents and manage compliance across a wide range of business purpose lending scenarios. With DossDocs, lenders can generate both initial disclosures, broker and state-specific disclosures, and closing documents within a single workflow, covering the full document process in one place. "The next wave of lending technology isn't about digitizing paper - it's about eliminating manual work with automated execution and human review," said Sourabh Chirimar, CEO at The Mortgage Office. "DossDocs' automated approach to compliance and disclosures fits exactly where we're investing: a platform where the technology does the heavy lifting so our customers can focus on lending." "This integration is the whole enchilada," said Dennis Doss, Founding Manager of DossDocs. "You get The Mortgage Office, one of the strongest LOS platforms out there, and DossDocs with reliable loan doc coverage built for the private lending industry." DossDocs introduces a usage-based approach to document generation, enabling lenders to produce complete document packages without setup fees or long-term commitments. TMO customers can leverage the DossDocs integration to: - Generate initial disclosures and closing documents together in a single process - Support 50-state compliant lending with real-time regulatory updates - Automate key calculations including APR, loan estimates, and disclosure requirements - Utilize a guided questionnaire designed for ease of use across teams - Incorporate broker, investor, and state-specific disclosures where applicable The integration is designed to complement existing document processes within TMO, giving lenders the ability to choose the right tools based on deal complexity, compliance needs, and business preferences. As part of TMO's broader ecosystem of integrations, this partnership reflects the company's continued focus on delivering flexible, purpose-built solutions that help lenders streamline operations, reduce manual work, and scale with confidence. Click here to learn more about the TMO and DossDocs integration. About The Mortgage Office The Mortgage Office (TMO) is the most trusted loan management platform empowering 1,100+ lenders around the globe. Its cloud-based platform streamlines operations through accurate, compliant, and automated solutions for loan origination, loan servicing, and fund management. With cutting-edge features like borrower and investor portals, construction draw management, third-party integrations, and numerous out-of-the-box reports, TMO enables you to modernize your lending operations and confidently manage even the most complex loans. Learn more at www.themortgageoffice.com/. About DossDocs DossDocs is a leading cloud-based platform developed by seasoned mortgage law attorneys with deep expertise in the complexities of lending. Offering best-in-class, 50-state compliant loan documents, DossDocs simplifies the entire loan documentation process with on-demand automation. No contracts, subscriptions, or onboarding required. Unlike other platforms, DossDocs provides instant access to compliant loan documents with no setup fees or upfront costs, allowing lenders to create compliant loan documents quickly and easily. For more information, visit www.dossdocs.com. The Mortgage Office. The Mortgage Office (TMO) has been the trusted choice of high-performance lenders for over four decades because Applied Business Software, Inc. understand what's at stake when you're managing millions in loans and vital investor relationships.

MCP for lenders: connecting your LOS to AI safely. MCP (Model Context Protocol) is an open standard that lets AI tools like Claude or ChatGPT securely read from and act on your systems, including a loan origination system, through a controlled server you govern, instead of copy-pasted exports and uncontrolled data flows. Lending software vendors have started shipping MCP servers natively, which shifts the question for lenders from "should we connect AI to our loan data" to "how do we do it without creating a compliance problem." This post answers the second question, from the intersection of lending and AI engineering. What is MCP, in lending terms? MCP is a standard way to put a governed interface in front of a system so AI tools can use it safely. In lending terms, it is how an AI assistant reaches your LOS or servicing system without anyone exporting data to a spreadsheet first. Strip away the generic technology framing and it is simple. Today, when someone wants an AI tool to help with loan data, the usual path is to export records and paste them into a chat window: uncontrolled, unlogged, and a compliance headache. An MCP server replaces that with a defined interface. The AI tool connects to the server, the server decides what it can see and do, and every interaction runs through something you own and can audit. The data stays behind your controls instead of leaving them. The reason this matters now is that the vendors are moving. The Mortgage Office announced an MCP server integration allowing customers to connect their lending data to external AI tools, alongside a rebuild of its origination platform with AI embedded (The Mortgage Office, 2026). When origination platforms ship this natively, MCP stops being a niche engineering idea and becomes a standard capability lenders are expected to use well. What can a lender actually do with an MCP-connected LOS? Ask questions of the whole portfolio in plain language, analyse loan tapes, surface exceptions, and draft reporting, without an export step or a dedicated analyst for every query. The work that used to require a data pull and a specialist becomes a direct conversation with the data. The concrete workflows are already visible in the market: lenders are already building data lakes from servicing data and using AI tools like Claude to analyze portfolios, surface trends, and produce reporting that previously required a dedicated analyst (The Mortgage Office, 2026). Translate that into a day's work and it means a servicing lead can ask which loans in a segment are drifting, an ops manager can get a first draft of a board report from live data, and an analyst can spend time on judgment instead of assembling the numbers. The loan tape is the natural anchor for this. So much of lending analysis reconciles to the tape, and an MCP-connected system lets you interrogate it conversationally instead of exporting and pivoting; if the concept is unfamiliar, its explainer on what a loan tape is covers the fundamentals. It is the same idea Codiot built into DataTalk, its conversational analytics work: plain-language questions against live data, with the reasoning shown, so teams stop waiting on reports. Why is "safely" the operative word? Because an AI tool with access to loan data is an access-control and audit problem before it is an AI problem. Permissions, read-versus-write boundaries, PII exposure, and logging decide whether the connection is an asset or a liability. The risks are specific and worth naming without drama. Permissions scoping determines what the AI can reach; a server that exposes the whole database because that was easiest is the classic mistake. Read versus write is the sharpest line: reading data to analyse it is low-risk, while allowing the AI to change records is a different category that needs deliberate controls. PII exposure has to be handled at the server, deciding what sensitive fields are even reachable and how they are masked. Audit trails matter because a regulated lender must be able to answer, after the fact, what was accessed and by whom. And how the model handles the data it sees, retention, training, and residency, is a question you ask before connecting, not after. None of this is a reason to avoid MCP. It is the reason to implement it as a governed interface rather than an open door. The protocol gives you the controls; using them is the work. Should you build an MCP server or wait for your vendor? If your LOS vendor ships an MCP server, start there. Build a custom one when you need cross-system scope or your vendor is lagging. The decision is about coverage and timing, not ideology. The honest framework is short. When your origination or servicing vendor already offers an MCP server, the fastest safe path is to turn it on and configure it well, because the hard integration work is done and your effort goes into scoping and governance. When your need spans multiple systems the vendor does not cover, or when your vendor has not shipped anything and you cannot wait, a custom MCP server built over your existing APIs is the answer, at the cost of a real engineering project. Codiot build these, both the custom servers and the safe rollout around a vendor's, so the choice can be made on your actual situation rather than on what is easiest to sell. The point is to match the build to the gap: do not custom-build what your vendor already ships, and do not wait indefinitely for a vendor that is not moving. What does a safe lender MCP rollout look like? Read-only first, scoped to specific data, with the outputs evaluated and every interaction logged, and only then scoped write actions. It mirrors how a careful team ships any agent: prove behaviour before granting autonomy. Here is the phased shape that keeps a rollout out of trouble: * Read-only to start. The AI can query and analyse, nothing more. Most of the value, portfolio queries, tape analysis, reporting, lives here and carries the least risk. * Scoped data. Expose the specific datasets the use case needs, not the whole system. Narrow is safe and easy to widen; wide is hard to walk back. * Evaluate the outputs. Test that the answers are correct, that the tool refuses when it should, and that it does not confidently invent. This is the same discipline its post on how to evaluate LLM outputs describes, applied to loan data. * Audit logging. Every access and action recorded, so the compliance answer exists before anyone asks the question. * Then, scoped write actions. Only once reading is proven do you enable the AI to take actions, behind clear permissions and full logging. That progression is deliberately the same one Codiot apply to AI agents and to Agentforce work: earn autonomy with evidence. For the system of record underneath all of this, its loan origination system solution work is where the LOS side is built to be interrogated safely in the first place. Where does this go next? Toward agents that do not just read loan data but act on it, with the governance bar rising as the autonomy does. The direction is clear even if the timeline is not. The current wave is analysis: asking questions, surfacing trends, drafting reports. The next is action: agents that draft decisions, prepare files, and update records within defined boundaries. The capability is arriving fast, and the credit side shows the pace, Accenture's 2026 Banking Technology Trends report documents AI-first credit systems increasing automated approvals by roughly 50% and decisioning throughput by 70-90% (Accenture, 2026). As AI moves from reading to acting on loan data, the safety work described above stops being optional and becomes the thing that decides whether a lender can use any of this at all. The lenders who set up governed, audited, read-first connections now are the ones who will be able to grant more autonomy later without flinching. If you want that foundation built properly, talk to Codiot. Faq. Is MCP secure enough for regulated lending data? The protocol enables control; the security depends entirely on the implementation. MCP itself is a way to expose systems to AI through a server you govern, which means you decide what data is reachable, what actions are permitted, and what is logged. That is the foundation for a secure setup, but it does not create one on its own. A poorly scoped MCP server is as risky as any other over-permissioned integration. Done properly, with read-only defaults, scoped data, and audit trails, it is more controllable than the copy-paste exports it replaces. Does MCP work with any LOS? What does an MCP integration cost? MCP vs API integration, what is the difference? Can AI change records through MCP?