Dominick Prevete: DSCR Lender Data 远程 MCP 服务器是什么?
Dominick Prevete: DSCR Lender Data 的远程 MCP 服务器是一个托管的 Model Context Protocol 端点,地址是 https://dominickprevete.com/api/mcp,AI 助手无需在本地安装或运行任何东西即可连接。

@Dominick Prevete
基本信息
HMDA investor-lender data, DSCR glossary, loan programs and indicative rates. Business-purpose only.
https://dominickprevete.com/api/mcpclaude mcp add dominick-prevete-dscr-lender-data --transport http https://dominickprevete.com/api/mcpTop investment-property lenders in one of the 26 covered states, from 2025 HMDA loan-level data: lender name, LEI, loan count, volume, average loan and purchase share, plus state totals and the retrieval date. Read-only; figures are the filing record, not a recommendation.
Every HMDA filer across the 26 covered states grouped by LEI and ranked by 2025 investor-property loan count, with the number of states each filer reported in, the universe totals and the top-25 concentration share. The same computation https://dominickprevete.com/investor-property-lenders renders. A 26-state aggregate, not a national one.
How the investor-lender data is built: the filter definition, the server-side versus local filtering split, the vintage and retrieval date, the covered states, and the stated limitations an agent should attach to any figure it reports.
Definitions from the DSCR and investor-loan glossary (26 terms). Pass a term to look it up (case-insensitive, partial matches accepted) or no argument to list every term. Definitions describe what a term is; none states a program parameter.
The National Loan Provider program set — dscr, dscr_cash_out, short_term_rental, fix_and_flip, ground_up, rental_portfolio, bridge, commercial — each with the parameters its page already publishes, its starting rate from the site's single rate source with the as-of stamp, and the page URL. Where a page states no figure, the tool states no figure.
Indicative starting rate by program, from the site's single rate source, stamped with its as-of month (currently October 2026). These are the published ranges, not pricing on any deal.
The entity answer: legal name, business model (a mortgage brokerage that arranges and places financing and is not a lender), business-purpose scope and the owner-occupied exclusion, address, phone, the states with published investor-lending data, and the canonical Person @id for the founder.
A read-only MCP server for real estate investors researching DSCR and investment-property financing. It returns investor-lender rankings built from public HMDA data, by state across 15 states and in aggregate, along with the methodology behind them. It also provides a glossary of DSCR and investor-loan terms, an overview of available loan programs, and indicative rate ranges stamped with their as-of date. Every response carries the brokerage's disclosure. Financing is arranged for business-purpose, non-owner-occupied investment properties only; nothing returned is an offer, approval or commitment to lend.
Dominick Prevete: DSCR Lender Data 的远程 MCP 服务器是一个托管的 Model Context Protocol 端点,地址是 https://dominickprevete.com/api/mcp,AI 助手无需在本地安装或运行任何东西即可连接。
把端点 https://dominickprevete.com/api/mcp 添加到任意支持 MCP 的客户端,比如 Claude Code、Cursor 或 VS Code。本页的接入代码块可以一步配置好每个客户端。
不需要。Dominick Prevete: DSCR Lender Data 的 MCP 服务器不需要认证——直接用端点地址连接即可。
Dominick Prevete: DSCR Lender Data 暴露的是 Streamable HTTP 端点,这是远程 MCP 服务器普遍使用、主流客户端都支持的传输方式。
Email for long-term client relationships: ARVX reads each person first and remembers every pair.
The 9apes MCP server lets an AI assistant read the GitHub Actions data 9apes already collects for your organization: runs, jobs, steps, and their durations. Ask it why a build got slower, which jobs take the longest, or
Search and read current AWS documentation, check regional availability and load AWS agent skills, with no AWS account needed.
Hosted AdLoop: Google Ads, Reddit Ads, GA4 and Tag Manager in claude.ai, ChatGPT and more, with a preview before every change.
Pay-per-use data and tools for AI agents: crypto token safety and rug checks, web search with cited answers, page-to-Markdown, up-to-date library docs, package safety, company research and UK/EU open data. 160 tools, 20
评论