Skip to main content
Glama
README.md
# frc-mcp — FinanceRateCalc MCP Server

[![financeratecalc/frc-mcp MCP server](https://glama.ai/mcp/servers/financeratecalc/frc-mcp/badges/score.svg)](https://glama.ai/mcp/servers/financeratecalc/frc-mcp)
[![AllMCPs Verified](https://allmcps.com/api/badge/financeratecalc-frc-mcp)](https://allmcps.com/mcp/financeratecalc-frc-mcp?verify=f03cf9df-b2a6-4e4a-b914-a60161b7f75d)

Read-only statistics over the complete 2025 US HMDA mortgage record (1,187,606 FHA credit
decisions, reverse mortgages excluded) plus dated, machine-readable US mortgage rule tables,
served to AI agents over MCP. Every figure returns with its period, population and program
scope, a use contract, and a receipt that can be verified later.

**The red line:** every figure is a historical aggregate. This server never accepts borrower
details and never returns an individual approval prediction.

## Remote endpoint (no install, no key)

```
https://frc-mcp.ziyetis.workers.dev
```

Streamable HTTP, stateless. MCP protocol revisions `2026-07-28`, `2025-11-25`, `2025-06-18`,
`2025-03-26`. Server version **1.57.0** (tools, resources and prompts; `get_official_figure` returns figures from official publications restated as printed with the document's SHA-256 and revision status — Level-2 receipts; A2A agent card at `/.well-known/agent-card.json`; a `citation` envelope on every figure-bearing result; every claim document carries a keyless CFPB Data Browser `reproduce` query). Listed in the official MCP registry as
`io.github.financeratecalc/frc-mcp`.

Client configs (Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Codex, Gemini CLI), an
instruction block for `CLAUDE.md` / `AGENTS.md`, and a first task to try:
https://financeratecalc.com/use-frc.html

## Tools (remote endpoint, 19)

| Tool | What it answers |
|---|---|
| `check_claim_contract` | Checks a sentence you intend to write against the figure's use contract; fixed patterns, no model. Returns `pass` / `needs_qualifier` / `block`, a safe wording, and a receipt on pass. Also checks sentences against the five rule tables, binding each figure to the role named in its clause. |
| `get_national_fha_stats` | 2025 national FHA denial rate with its universe. |
| `get_lender_denial_stats` | One of the 100 largest FHA lenders: volume, denial rate, rank, applicant-mix caveat. |
| `list_lenders` | The 100 largest FHA lenders ranked by denial rate or volume. |
| `get_state_denial_stats` | State FHA denial rate and small-loan split. |
| `get_metro_lender_gap` | Within one metro, the spread between the highest- and lowest-denying high-volume lenders. |
| `get_small_loan_penalty` | Denial rate under $150K vs over $250K, by state. |
| `get_denial_reason_shares` | Median and maximum share of each HMDA denial reason across the top-100 lenders. |
| `get_door_effect_summary` | Lender identity's share of explained variation in 2025 FHA denials. |
| `get_application_outcomes` | Where FHA applications end: originated, approved-not-accepted, denied, withdrawn, closed-incomplete as shares of applications acted on, nationally or for one lender, with the reconciliation to the decisioned denial rate. |
| `locate_denominator` | Which definition reproduces a quoted 2025 denial rate: program (FHA, VA) x loan purpose x scope x base, with counts. A match is not an attribution. |
| `get_official_figure` | A figure from an official publication (HUD's FHA production report and FHA loan performance report, Ginnie Mae's issuance summary, SBA's 7(a) and 504 activity file) exactly as printed, with the document's SHA-256, table and row, period, exclusion note, revision status and a receipt; nothing computed. Params: source, period (YYYY-MM), figure. |
| `get_conditional_door_map` | Per-lender conditional surface with its robustness record. |
| `list_cohorts` / `run_cohort` | Ready-made public cohorts; no upload needed. |
| `get_rule` | One of five dated rule tables: FHA annual MIP (ML 2023-05), FHA forward loan limits 2026, FHFA conforming limits 2026, CFPB HOEPA/QM thresholds 2026, VA funding fee. |
| `resolve_fha_annual_mip` | The MIP cell for a stated term, base loan amount and LTV; asks for a missing condition instead of guessing it. |
| `screen_counterparties` / `screen_counterparties_quote` | Counterparty screen; free preview, full brief needs a licence bought by a human. The agent only relays the offer. |

Plain HTTP beside MCP: `GET /server-card`, `GET /cite/<claim-id>` (receipted sentence in text,
Markdown, HTML and BibTeX), `GET /verify?r=<receipt>` (is this figure still current; if not,
which correction superseded it).

## Receipts

A figure travels with a receipt such as
`⟦FRC:national-fha-denial-rate-2025:22.1%:ebf6b00b:m2610⟧`: claim id, value, the first eight
hex characters of a hash of the claim object, and the channel and month it was issued through.
`/verify` answers `current`, `stale` (corrected since; the correction is returned),
`hash-current-value-mismatch` (the quoted value was edited) or `unrecognized` (this publisher
never issued it). Conformance runs on every deploy and is published:
https://financeratecalc.com/conformance.html

## Local stdio package (npm)

The `frc-mcp` npm package is the original five read-only tools from a bundled data snapshot; it
does not carry the rule tables, the sentence gate or receipts. Prefer the remote endpoint.

```json
{
  "mcpServers": {
    "financeratecalc": {
      "command": "npx",
      "args": ["-y", "frc-mcp"]
    }
  }
}
```

## Source

The remote server's source of record is `worker/index.js` in
https://github.com/financeratecalc/financeratecalc.github.io, deployed to Cloudflare Workers
on every change to that file. `remote/worker.js` in this repository is an early (five-tool)
copy kept for reference.

Data: CC BY 4.0, attribution to financeratecalc.com. Source of truth: public CFPB HMDA 2025
(loan_type 2; actions 1, 2, 3; denial = action 3). Method: https://financeratecalc.com/methodology.html
· corrections log: https://financeratecalc.com/corrections.html · SSRN 7156938.
No lender recommendations, no individual predictions, no lender pays for placement.

TDQS

A4.1/5.0

Scored across 5 tools

Disambiguation4/5

Each tool targets a distinct scope: national, lender-specific, state-level, lender listing, and door-effect analysis. There is minor overlap between list_lenders and the lender tables inside get_door_effect_summary, but the descriptions make the intended use clear.

Naming Consistency4/5

Most tools follow a get_<scope>_stats pattern, which is predictable and readable. list_lenders and get_door_effect_summary deviate slightly from that pattern, but the naming remains consistent in style and verb usage.

Tool Count5/5

Five tools is well-scoped for a specialized FHA denial statistics server. Each tool covers a meaningful slice of the domain without redundancy or bloat.

Completeness5/5

The tool surface covers national aggregates, state-level views, individual lender stats, lender rankings, and the cross-lender door effect analysis. For the stated purpose of exposing FHA denial statistics, there are no obvious missing operations or dead ends.

Maintenance

ActivityActive
ResponsivenessLimited data