Cal MCP Server

Cal MCP Server

Enables AI assistants to answer mortgage-related queries by providing tools for lender search, loan limit lookup, down-payment assistance programs, and more, with data sourced from real wholesale lenders and broker-curated intel.

Category
Visit Server

README

Cal MCP Server

The mortgage MCP server, with a real lender data moat. 166+ wholesale lenders · 1,400+ indexed program documents · 250+ down‑payment‑assistance programs · 50 states. First Model Context Protocol server in the mortgage industry with broker‑accessible matrix data behind it — not a calculator, not a rate scraper.

@homeloanexpress/cal-mcp-server exposes Cal — the AI mortgage platform from HomeLoanExpress — to any Model Context Protocol client: Claude Desktop, Cursor, Continue, Goose, Zed, Cody, your own agent. Loan officers, brokers, real‑estate agents, and consumer borrowers can ask their AI assistant questions like:

  • "What's the 2026 conforming loan limit in Alameda County, CA, for a 1‑unit?"
  • "Top 5 wholesale jumbo lenders in California for a 712‑FICO, 80% LTV, $1.4M loan with a departing residence still listed."
  • "What DPA programs in Orange County California work for a first‑time buyer with 700 FICO?"
  • "Plaza Home Mortgage — what's their current master credit guideline document, and which AE covers Northern California?"
  • "Show me the structuring playbook for a sub‑580 FHA file."

…and Cal returns answers grounded in current matrix PDFs, FHFA / HUD / VA source documents, and tribal‑knowledge intel curated by working brokers.

Why this matters

The mortgage industry has spent a decade in the "everybody buys Encompass" era. AI hasn't touched the workflow because the source data is gated, fragmented, and expires. Cal's library — built from broker‑portal scrapes, indexed weekly, kept current — is the data layer that makes any mortgage AI actually useful. This MCP server is how that data reaches every agent platform built on top of MCP.

If you're an LO, this means asking Claude (or Cursor, or your own agent) the same questions you used to dig through PDFs to answer — and getting an answer in seconds with a source citation.

Tools

The server exposes seven tools:

Name Purpose
cal_fact_lookup FHFA conforming loan limits, FHA loan limits, VA funding fee, FHA MIP, agency DTI caps. County‑level. 50 states + DC.
cal_search_lenders Scenario‑aware wholesale‑lender finder. Pass FICO / LTV / loan amount / state / occupancy → ranked top 5 (or top 15 with expand=true) with score, qualifying reasons, blockers.
cal_lender_details Full record for one lender — programs, state licenses, AE contact, broker portal URL, status, recent matrix dates.
cal_lender_documents List indexed matrices / handbooks / overlays for a lender. Sorted master → current → superseded. Each doc includes a short extracted summary.
cal_dpa_search Down‑payment‑assistance programs by state / county / city / FTHB / type. Up to 25 ranked programs with provider URLs.
cal_lender_intel Tribal‑knowledge intel — strengths, watch‑outs, speed notes, scenarios they love, scenarios to avoid, AE contacts. Pass intent to narrow.
cal_scenario_pattern Structuring playbooks for hard scenarios: departing residence, ITIN / foreign national, sub‑580 FHA, renovation, super‑jumbo, plus the matrix‑freshness citation rule.

The tool surface mirrors Cal's internal tool‑use schema, so any model (Claude Sonnet, Opus, Haiku, GPT‑4o, Gemini, …) that calls them gets the same shape Cal itself uses.

Install

Claude Desktop (macOS / Windows)

Add this to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "cal": {
      "command": "npx",
      "args": ["-y", "@homeloanexpress/cal-mcp-server"],
      "env": {
        "CAL_API_TOKEN": "your-token-here"
      }
    }
  }
}

Restart Claude Desktop. The cal tools appear in the slash menu.

Cursor

Settings → MCP → Add Server, then paste the same JSON.

Continue / VS Code

~/.continue/config.jsonmcpServers block, same shape.

Local dev (clone + link)

git clone https://github.com/homeloanexpress/cal-mcp-server.git
cd cal-mcp-server
npm install
npm run build
# Then point your MCP client at dist/index.js with `node` as the command
# and an absolute path in args[0].

Getting a token

CAL_API_TOKEN is an opaque bearer token. Use a long‑lived API key, not a session token — session tokens from /auth/login expire after 8 hours, which means an MCP client config that worked today returns 401 tomorrow.

Recommended — long‑lived API key (cal_live_…)

Sign up at vault.homeloanexpress.ai (request access if you don't have a key yet — broker LOs are auto‑approved). Then:

# 1. Get a short‑lived session token from /auth/login
SESSION=$(curl -s -X POST https://vault.homeloanexpress.ai/auth/login \
  -H "Content-Type: application/json" \
  -H "Origin: https://homeloanexpress.ai" \
  -d '{"email":"<your-email>","password":"<your-password>","portal":"team"}' \
  | sed -n 's/.*"token":"\([^"]*\)".*/\1/p')

# 2. Mint a long‑lived API key
curl -X POST https://vault.homeloanexpress.ai/auth/api-keys \
  -H "Content-Type: application/json" \
  -H "Origin: https://homeloanexpress.ai" \
  -H "Authorization: Bearer $SESSION" \
  -d '{"label":"my-claude-desktop"}'

The response includes a key field starting with cal_live_. Copy it now — it's only shown once. Use it as CAL_API_TOKEN. The key has no expiry; revoke it any time with DELETE /auth/api-keys/<id> (list yours with GET /auth/api-keys).

Fallback — session token

If you just want to try a tool call quickly, the session token from /auth/login works too (8‑hour TTL). Don't put one in a long‑lived MCP config.

v0.2 preview

OAuth 2.1 with PKCE is on the v0.2 roadmap so MCP clients can run the consent flow themselves. Until then, the API key flow above is the recommended path.

Verify it works

After installing, ask your MCP client:

"What's the 2026 conforming loan limit in Alameda County CA?"

You should get back something like:

$1,229,100 (FHFA 2026 Conforming Loan Limits, announced 2025‑11‑26).
High‑balance HCOL county. Anything above this is a jumbo loan.

If you don't, check:

  1. The token isn't expired — Cal session tokens last 24 hours by default. Refresh via /auth/login.
  2. Network can reach https://vault.homeloanexpress.ai.
  3. Your MCP client picked up the config — restart it after editing.

Configuration

Env var Required Default Notes
CAL_API_TOKEN yes Bearer token from /auth/login.
CAL_API_BASE_URL no https://vault.homeloanexpress.ai Override only if you're testing against a Cal instance other than production.

Data scope

Calls are scoped to the list tied to your token. If you're a broker LO with a curated approved‑lender list, results are filtered to lenders you can actually use. Cal will not leak across approved lists. The chris-hle demo list (the founder's list) is the broadest at 166 lenders.

License

MIT. The wrapper code is open. Cal's curated lender library, DPA program data, and lender intel are proprietary and remain behind the API.

About Cal

Cal is the AI mortgage platform from HomeLoanExpress, built by a 17‑year industry veteran (NMLS 275073) for working brokers. Production state at this release:

  • 166 wholesale lenders, 1,424 indexed program documents, 250 DPA programs, 50 states
  • 4 LOs in pilot, 342 questions answered in the first 24 hours of soft‑launch
  • $500K committed + $1.5M soft‑committed → $2M seed target

Investor preview at vault.homeloanexpress.ai/pitch.html.

Roadmap

  • v0.1 (current) — 7 tools, API‑key auth, stdio transport, npm + GitHub.
  • v0.2 — OAuth 2.1 with PKCE. .dxt Claude Desktop bundle. Optimal Blue MCP companion server.
  • v0.3 — Streamable HTTP transport. Public agency MCPs (Fannie Selling Guide, Freddie Guide, FHA Handbook 4000.1, VA Handbook 26‑7, FHFA limits, HUD AMI) under the same org as open‑source siblings.
  • v0.4 — Encompass MCP companion (LOS read/write). DocuSign MCP companion. The investor‑deck card 8 vision: one prompt fires across CRM + email + calendar + pricing + library.

Contributing

Issues and PRs welcome. The roadmap above is the priority order; if you want to take a piece, open an issue first so we don't dual‑track.

Contact

Chris Black · NMLS 275073 · chris@homeloanexpress.ai · 925‑286‑7681

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured