Question 1

What is the Model Context Protocol (MCP)?

Accepted Answer

MCP (Model Context Protocol) is an open standard for exposing tools and data sources to AI agents, created by Anthropic. Instead of building proprietary integrations for each AI system, MCP defines a universal interface — like USB for AI tools. Any AI agent that speaks MCP can discover and invoke Confer's 32+ mortgage tools through a standardized API, without requiring vendor-specific SDKs or proprietary plugins. Think of it as an open architecture that prevents vendor lock-in while enabling any LLM (GPT-4, Claude, Gemini, LLaMA, etc.) to perform mortgage underwriting, document analysis, and compliance checks through Confer's platform.

Question 2

Why does open architecture matter for mortgage lenders?

Accepted Answer

Open architecture prevents vendor lock-in and future-proofs your technology investment. With Confer's MCP platform, you're not tied to a single AI vendor or forced to rebuild integrations when new models emerge. If OpenAI releases GPT-5 tomorrow or your compliance team prefers Claude for certain tasks, those models can immediately leverage Confer's 32 MCP tools without custom integration work. Contrast this with closed SDK platforms like Encompass, where adding a new AI capability requires vendor approval, lengthy certification cycles, and proprietary API agreements. Open architecture also enables competitive pricing — you can switch LLM providers based on cost, performance, or regulatory requirements without rebuilding your mortgage intelligence layer.

Question 3

What MCP tools does Confer expose?

Accepted Answer

Confer exposes 32+ MCP tools across four domains: (1) Underwriting — 14 tools including credit analysis, DTI calculation, LTV/CLTV validation, eligibility checks, condition generation, and full underwriting analysis. (2) Document AI — 8 tools covering classification, data extraction, cross-reference verification, and automated E-Folder assignment. (3) Compliance — 10 tools for QM/ATR validation, TRID timer management, HMDA field population, HOEPA/HPML checks, and regulatory audit trails. (4) Communication — borrower notifications, agent messaging, escalation routing, and voice AI handoff. Every tool includes structured inputs, typed outputs, and error handling.

Question 4

How is Confer's MCP platform different from Encompass APIs?

Accepted Answer

Encompass uses a closed, proprietary SDK that requires vendor partnership agreements, certification processes, and platform-specific integration code. Only approved vendors can build Encompass integrations, and developers must learn ICE's proprietary API architecture. Confer's MCP platform is open — any AI agent or development team can discover available tools, read their schemas, and invoke them through a standardized protocol without vendor approval. This means faster innovation, lower integration costs, and the ability to use best-of-breed AI models instead of being locked into a single vendor's roadmap. Additionally, MCP is an emerging industry standard backed by Anthropic, meaning tools built on MCP today will work with tomorrow's AI systems.

Question 5

Can I use my own AI models with Confer's MCP tools?

Accepted Answer

Yes. Confer's MCP platform is model-agnostic. If you've fine-tuned a custom LLM on your institution's historical loan data, that model can invoke Confer's underwriting, document AI, and compliance tools just like GPT-4 or Claude would. You can also run entirely local models (LLaMA, Mistral, etc.) for sensitive operations while using cloud-based models for routine tasks. The MCP protocol doesn't care what AI is making the request — it only cares that the request format matches the tool schema. This flexibility is critical for regulated industries where data residency, audit requirements, or competitive advantages demand control over the AI layer.

Question 6

What does implementation look like for MCP integration?

Accepted Answer

MCP integration typically takes 2–4 weeks depending on your AI infrastructure. Confer provides tool discovery endpoints, schema documentation, and sample code for invoking each tool. If you're using an MCP-native AI framework (like Claude Desktop, LangChain with MCP support, or custom agents built on Anthropic's MCP SDK), integration is straightforward — point your agent at Confer's MCP server, and tools auto-register. For legacy systems or custom AI platforms, Confer's engineering team can assist with adapter development. Once integrated, new tools added to Confer's platform automatically become available to your AI agents without additional deployment work.

Question 7

Is MCP a security risk? Who controls access to these tools?

Accepted Answer

MCP tools are protected by the same enterprise security controls as Confer's web application: OAuth 2.0 authentication, role-based access control, request-level authorization, and full audit logging. An AI agent can only invoke tools that the authenticated user (or service account) has permission to access. For example, a borrower-facing AI chatbot cannot call underwriting decision tools, and a loan officer's AI assistant cannot access underwriter-only functions. Every tool invocation is logged with timestamp, user identity, input parameters, and output results — providing complete audit trails for regulatory review. MCP doesn't bypass security — it inherits it.

Feature	Confer (MCP)	Encompass (Closed SDK)	Others
Protocol standard	MCP (open, Anthropic-backed)	Proprietary SDK	No standard
Vendor approval required	✗ Open access	✓ Partnership required	N/A
Model flexibility	Any LLM (GPT, Claude, Gemini, local)	Vendor-controlled only	Limited
Tool discovery	✓ Auto-discovery via schema	Manual documentation	Manual
Integration timeline	2–4 weeks	6–12 months (certification)	Varies
Cost to integrate	Open (no licensing)	Partner fees + certification	Custom dev
Future-proof	✓ New models work immediately	✗ Vendor roadmap dependent	✗
Underwriting tools	14 MCP-native	Closed API	None
Document AI tools	8 MCP-native	Third-party required	None
Compliance tools	10 MCP-native	Limited native support	None
Audit logging	✓ Every invocation logged	✓ (proprietary format)	Varies
Multi-model orchestration	✓ Use different LLMs per task	✗	✗

The First Mortgage PlatformBuilt on the Model Context Protocol

Why Open Architecture Matters

Closed SDK (Encompass)

Open MCP (Confer)

Four MCP Tool Domains

Underwriting Tools

Document AI Tools

Compliance Tools

Communication Tools

MCP vs. Closed SDK Comparison

Frequently Asked Questions