How to Use the Lender Price MCP to Run Rate Comparisons for Loan Officers

A loan officer running 30 conversations a week spends maybe 8 of those hours just on rate explanations. Buy-up vs. buy-down. Lender credits vs. discount points. ARM vs. fixed at this LTV. The math is in the LOS. The narrative is in your head. The thing that ships to the borrower is whatever Outlook gives you 90 seconds before your next call.

This post wires those three pieces together: the Lender Price pricing engine via MCP, the Loan Officer AI Cowork Vault for the compliance-aware narrative, and Claude as the layer that holds the conversation in your voice — TILA, RESPA, and ECOA guards running ambient on every draft.

💡 The stack. Lender Price MCP supplies the pricing scenarios. The Loan Officer Vault is the agentic layer that turns those scenarios into borrower-facing comparisons in your voice, with always-on TILA/RESPA/ECOA guards. $19 one-time, both Claude Cowork and Microsoft 365 Copilot plugins in one purchase.

What the Lender Price MCP actually does

Lender Price is the pricing engine the mid-market and wholesale mortgage world runs on — investor product matrices, real-time pricing, eligibility checks, LLPA overlays. Their REST API exposes those endpoints by loan scenario; the MCP server is the bridge that lets Claude call them inside a conversation.

What the MCP returns when you ask for a rate scenario:

• Base rate at requested LTV, FICO, loan amount, occupancy, doc type
• LLPA adjustments applied (with named line items, not a black-box single number)
• Lender credits / discount points schedule at par, +0.25%, +0.50%
• Eligibility flags (DSCR, jumbo limits, state-level overlays)
• Comparable scenarios — what changes if you bump LTV by 5%, change DTI tier, switch from primary to second home

It does NOT return borrower PII. Scenarios are anonymized by design — that's a design choice that matches how compliance auditors expect pricing logs to look.

Pre-flight: what you need installed

• Claude Cowork (desktop) or Claude Code — current versions support MCP servers
• A Lender Price API key — from your wholesale or correspondent account; check with your account exec if you don't have one
• Node.js 20+ if running the community MCP locally (or hosted-MCP if your firm runs its own)
• The Loan Officer AI Cowork Vault installed in your Cowork Project — that's what the pricing feeds into

Configure the MCP server

Open Claude's config:

{
  "mcpServers": {
    "lender-price": {
      "command": "npx",
      "args": ["-y", "@lenderprice/mcp-server"],
      "env": {
        "LENDERPRICE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Restart Claude. Confirm under Customize → Connectors that the server is connected. If it's not, the most common issue is a stale API key or a Node version below 20.

Run a comparison pull

Inside your Cowork Project (with the vault already installed):

Use the Lender Price MCP to pull pricing for:
- Conventional 30-year fixed
- $425,000 loan amount
- 80% LTV
- 740 FICO
- Primary residence, single-family
- Standard doc type
- 30-day lock

Show base rate + par, +0.25 points, +0.5 points, and break out the LLPAs.

Lender Price returns the matrix. Claude formats it into a readable table. You eyeball the LLPAs to confirm nothing weird (e.g., a state overlay you forgot).

Chain it with the vault for the borrower-facing version

Raw pricing is for you. The borrower needs the narrative. Call the vault skill:

/rate-comparison-explainer
Borrower: [Borrower First Name]
Loan scenario: 30-year fixed, $425k, 80% LTV, 740 FICO, primary
Pricing scenarios: [paste Lender Price table, or say "use the pull above"]
Recommended scenario: par rate, no points
Why this recommendation: short break-even on points given borrower's planned hold
Tone: warm, plain-English, no jargon
Compliance: TILA/RESPA aware

The vault produces a borrower-ready email. The TILA disclosure language is in. The "APR may differ from note rate" line is in. The "subject to credit approval and final underwriting" is in. The realtor isn't CC'd by accident (the vault asks before CC).

What you get back is a 250-word email in your voice that the borrower will actually read. No screenshot of a pricing engine UI, no PDF of LLPAs they can't parse.

Two compliance guards firing in the background

While you're drafting, the vault's ambient guards are running:

• TILA/RESPA guard — flags drafts that promise a specific APR without the "subject to" qualifier, or that compare loan products without the required side-by-side disclosure language
• Fair-lending (ECOA) guard — flags language that could be read as steering based on protected class. "This program is great for first-time buyers like you" is fine. "This program is great for young families" is not. The guard catches the second before it ships.

You don't invoke either guard. They invoke themselves when the draft drifts into territory that needs the check.

Chain into the realtor co-marketing piece

The same pricing context can seed a weekly realtor email:

/realtor-rate-update
Realtor: [Realtor First Name + Brokerage]
Audience: their buyer list this week
Pricing scenarios: [from the Lender Price pull above]
Note: include conventional + FHA + a jumbo example
Tone: confident, useful, not promotional
Compliance: RESPA Section 8 aware (no kickbacks, no co-marketing fee implied)

The vault produces a weekly rate-summary email the realtor can forward to their buyer pipeline. RESPA Section 8 considerations (no implied co-marketing fees) are handled by the same ambient guard that fires on borrower-facing material.

Common failure modes

• Pricing comes back with stale rates. Lender Price pulls live, but if your firm runs cached pricing for performance, you may be seeing yesterday's rates. The MCP has a lock_in_window flag — set it to the actual lock window you'd offer the borrower so the math doesn't diverge.
• LLPA overlays surprise you. State-level overlays and investor-specific adjustments stack. The MCP returns them itemized — read the lines.
• The vault's draft sounds generic. Run /lo-setup-wizard if you haven't yet. The wizard captures your firm name, your designation (MLO vs. broker), your state(s), your NMLS ID, and 2–3 voice samples. Drafts after that point sound like you, not like a generic mortgage email.
• Fair-lending guard fires on a draft you thought was fine. Read the flag. It will name the specific phrase. If you disagree, you can override — but read the flag first; it's usually catching something you didn't intend.

When you'd skip this stack

• You close ≤ 4 loans a month. The MCP setup time isn't worth it. Use Lender Price directly via their web UI for that volume.
• Your firm has a pricing-tool monopoly clause. Some wholesale relationships restrict third-party API access. Check with your account exec before plumbing the MCP in.
• You don't run wholesale. Retail-only LOs on a single investor don't see enough scenario variation to benefit. The math becomes "what's the rate today" not "how do these five scenarios compare."

Sources

• Lender Price: Developer API docs (vendor primary)
• Anthropic: Model Context Protocol (MCP) introduction
• CFPB: TILA-RESPA Integrated Disclosure (TRID) rule resource
• CFPB: Equal Credit Opportunity Act (ECOA) compliance