underwriting-mcp
A headless MCP connector that lets you underwrite commercial and multifamily deals inside a Claude chat, providing exact arithmetic for NOI, valuation, debt sizing, DSCR, and returns.
README
Underwriting MCP Server
A headless Model Context Protocol connector that lets you underwrite commercial and multifamily deals inside a Claude chat. Claude runs the conversation; this server does the arithmetic that must be exact (NOI, valuation, debt sizing, DSCR, returns) and serves your firm's underwriting methodology so every conversation underwrites the same way.
There is no web UI. The interface is Claude with this connector attached.
This server does deterministic math + methodology only. It does not open, fill, or recalculate
Commercial_Template.xlsx— that's the Skills pipeline's job in code-execution. The two share the same underwriting constants so they never disagree.
Stack
- Next.js (App Router) +
mcp-handler— wraps the route as an MCP endpoint over stateless Streamable HTTP (no SSE, no Redis; free-tier friendly). - TypeScript + Zod for tool input contracts.
- Deploy target: Vercel (Hobby/free tier).
What it exposes
Tools (lib/calc/)
| Group | Tools |
|---|---|
| Common (§4.1) | calculate_noi, value_from_cap_rate, implied_cap_rate, size_debt, debt_metrics, cash_on_cash, equity_required, exit_value, run_scenarios, deal_summary |
| Commercial (§4.2) | summarize_rent_roll, market_rent_gap, nnn_reimbursement, rollover_exposure |
| Multifamily (§4.3) | summarize_unit_mix, economic_occupancy, per_unit_metrics, proforma_rent_growth, mf_expense_defaults |
Resources (lib/methodology/)
underwriting-conventions, scenario-definitions, commercial-field-map, multifamily-field-map,
template-version — served over underwriting://… URIs.
Prompts (lib/prompts/)
underwrite_retail_commercial, underwrite_multifamily — guided openers that collect the right
inputs, call the tools in order, and present the four-scenario grid.
Conventions (enforced)
- Four scenarios always: List / Target / Probable / Conservative. Per-scenario levers are cap rate, LTV, exit cap rate; everything else is shared.
- Rates are fractions —
0.055= 5.5%. The server rejects rate-like inputs> 1.0to catch the classic5-vs-0.05bug. - Stabilization year selector (2/3/4, default 4); any stabilized result states which year it used.
- Every denominator (cap rate, equity, loan, GPR) is guarded — clear error instead of
Infinity/NaN. - Multifamily defaults (overridable): mgmt fee 3% of EGI, replacement reserves $250/unit/yr, rent trend 3%/yr.
Develop
npm install
npm test # hand-checked deal per asset class (node:test via tsx)
npm run typecheck # tsc --noEmit
npm run dev # http://localhost:3000/api/mcp
Deploy (free tier)
npm test— confirm every formula matches its hand-checked deal.- Push to GitHub.
- Vercel → New Project → import repo → Deploy (Node auto-detected).
- Your endpoint is
https://<project>.vercel.app/api/mcp. - (Optional) set
MCP_API_KEYin Vercel env to requireAuthorization: Bearer <key>.
Add the connector in Claude
Requires a Pro, Max, Team, or Enterprise plan (custom connectors over remote MCP).
- Settings → Connectors → Add custom connector.
- Paste the
…/api/mcpURL; complete auth if you set a key. - Start a chat, confirm the tools/prompts appear, run the two walkthroughs.
Layout
app/api/mcp/route.ts # mcp-handler entry: registers tools, resources, prompts + shared-key auth
lib/calc/finance.ts # §5 primitives (debt, DSCR, value, exit)
lib/calc/common.ts # §4.1 common tools incl. run_scenarios
lib/calc/commercial.ts # §4.2 retail/office/net-lease
lib/calc/multifamily.ts# §4.3 multifamily
lib/validate.ts # fraction-vs-percent + zero-denominator guards
lib/methodology/ # §6 resources (source of truth, one file each)
lib/prompts/ # §7 guided flows
test/ # one hand-checked deal per asset class
Roadmap
- Phase 4 — hardening: OAuth via
withMcpAuth, rate limiting, multi-year levered cash flow / IRR (ARGUS-style) if desired.
Recommended Servers
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.