The lender list your assistant gave you last week was probably invented
Ask a general-purpose AI assistant which lenders write bridge debt on a four million dollar retail property in Georgia and it will answer. It will name firms, describe their boxes, and sound certain. Some of those firms will be real. Some of the criteria will be from 2021. At least one of them will not exist in the form described.
This is not the model being careless. It is the model doing the only thing available to it. There was no authoritative, machine-readable source of current CRE lender criteria for it to read, so it reconstructed one from whatever it absorbed during training. A reconstruction is not a directory.
As of this week, there is a source.
What we published
Finance Lobby now serves a full agent-discovery surface at the root of financelobby.com. In plain terms: an AI assistant, an agent, or any piece of software can find out what Finance Lobby knows, ask it questions, and get back current data with a date attached, without a human in the loop and without an account.
The pieces, and what each one is for:
A Model Context Protocol server, at financelobby.com/api/mcp. This is the part most people will actually use. MCP is the standard that lets an assistant call an outside tool mid-conversation. Nine tools are exposed:
| Tool | What it answers |
|---|---|
search_lenders | Which lenders fund this property type, loan size and state |
get_lender_profile | One lender's box, recent quotes and funded deals |
get_platform_stats | Platform-wide lending statistics |
search_open_deals | Which CRE deals are seeking financing right now |
get_benchmark_rates | Today's Treasury curve, SOFR, Prime, Fed Funds |
get_rate_history | The same series over 1M to 5Y, for charting |
list_states / list_property_types | Reference data, so filters are valid |
calculate_mortgage | Debt service, coverage and the constant |
Transport is streamable HTTP. Authentication is none. Every tool is public, read-only and free.
An agent skills index, at /.well-known/agent-skills/index.json, published against the Agent Skills Discovery RFC. Each skill is a markdown file with a SHA-256 digest, so an agent can verify it read what we published rather than a cached copy of it.
An API catalog at /.well-known/api-catalog, in the linkset format specified by RFC 9727. Nine APIs, each with a link to the OpenAPI spec and the documentation.
OAuth and OIDC discovery metadata, plus an auth.md at the site root, so an agent acting on behalf of a signed-in user can work out how to register and authenticate rather than guessing.
An ARD capability manifest at /.well-known/ai-catalog.json, with representative queries attached to each entry so registries can build semantic embeddings from them.
Payment protocol discovery. x402, MPP, UCP and ACP profiles are published. Nothing on the platform charges an agent today. The profiles exist so that the day something does, the plumbing is already standard rather than bespoke.
The line we drew, and why it matters more than the list above
Every tool is read-only. Not one of the nine can create a deal, send a message to a lender, accept a quote, change a preference, or spend a dollar.
That was a deliberate decision, and it is the part worth arguing about. The obvious next step for a platform publishing agent tools is to let agents act: submit the deal, accept the quote, sign the thing. We are not doing that yet, and the reason is that a CRE financing decision has a person's signature under it. An assistant that can research your options is a research assistant. An assistant that can accept a quote on your behalf is an agent in the legal sense, and that is a different product with different consent, disclosure and liability questions attached.
So: read the market, freely and without an account. Act on it as yourself.
What this actually changes for you
If you are a broker. Your assistant can now pull a lender shortlist that reflects current stated criteria rather than a 2021 memory, and it can pull today's Treasury curve into the same answer. You still shop the deal. You just start from a list that is real.
If you are a lender. Your directory profile is now readable by the tools your prospective counterparties use to search. The criteria you set on Finance Lobby are the criteria a machine reports. If your box has not been updated in a year, that is now visible in more places than it used to be. Drop your lending preferences and they propagate.
If you are a borrower. The benchmark rate data an assistant quotes you now comes from the Federal Reserve's own FRED series with a publication date attached, not from a blog post it read once. Ask it for the date. It has one.
How to point an assistant at it
Any MCP-capable client takes a server URL. Ours is:
https://financelobby.com/api/mcp
No key, no account, no sign-up. If you want to see the catalogue first, open financelobby.com/llms.txt in a browser. It is written for machines and is perfectly readable by people.
The honest caveat
Two of these standards are drafts, not finished specifications. WebMCP is an early browser proposal. The agent payment protocols are a year old at most and nobody has settled which of them wins. We published against all of them on purpose, because the cost of being early on a discovery file is one file, and the cost of being late is that the machines answering your customers' questions describe you from memory.
Frequently asked questions
Does Finance Lobby have an MCP server?
Yes. It is at https://financelobby.com/api/mcp, uses streamable HTTP transport, requires no authentication, and exposes nine read-only tools covering lender search, lender profiles, platform statistics, open deals, benchmark interest rates, rate history, reference data and a mortgage calculator.
Do I need an account or an API key to use the Finance Lobby agent tools?
No. All nine MCP tools are public, read-only and free. Authentication is only required for the protected APIs that act on a specific user's own deals, and those are not exposed as agent tools.
Can an AI agent post a deal or accept a quote on Finance Lobby?
No. Every exposed tool is read-only by design. No tool can create a deal, message a lender, accept a quote, change lending preferences or make a payment. Actions that commit a party to anything are done by a person.
Where does the Finance Lobby benchmark rate data come from?
From the Federal Reserve's FRED series, refreshed daily. Every value carries an asOfDate publication date and an isStale flag that is set when the last fetch failed and the figure shown is the previous good one.
What is an agent skills index?
A discovery file at /.well-known/agent-skills/index.json that lists the instruction documents a site publishes for AI agents, each with a name, description, URL and SHA-256 digest so the agent can verify integrity. Finance Lobby publishes one against the Agent Skills Discovery RFC.
Set your box once and let it answer for you.
Lenders on Finance Lobby state their criteria once and receive matched, quote-ready deal flow. Those same criteria are now what a machine reports when someone asks who funds a deal like yours.
Start for free