FinanceRateCalc (frc Mcp)

What FinanceRateCalc (frc-mcp) does

FinanceRateCalc (frc-mcp) exposes historical mortgage statistics and related rule data to AI agents through the Model Context Protocol. Its FHA analysis uses the public 2025 CFPB HMDA record, covering 1,187,606 FHA credit decisions and excluding reverse mortgages. The analysis defines the population as FHA loans with loan type 2 and action codes 1, 2, or 3; an action of 3 represents a denial.

The service is intended for aggregate research. It does not accept borrower details, produce an approval or denial prediction for an individual, or recommend a lender. Reported figures include their period, population, program scope, use conditions, and—on the remote service—a receipt that can be checked later.

How it works

The remote deployment runs as a stateless Streamable HTTP MCP service at https://frc-mcp.ziyetis.workers.dev. It provides the expanded publisher implementation, including read-only tools, resources, prompts, dated rule tables, claim checking, and receipt verification. Claims can also be retrieved through HTTP endpoints for citation and verification: /cite/<claim-id> returns a receipted sentence in several formats, while /verify?r=<receipt> checks whether a quoted figure remains current.

The local npm package runs over stdio from a bundled data snapshot. It contains the original five read-only analysis tools, but not the remote rule tables, sentence gate, or receipts. The package can be launched with npx -y frc-mcp from an MCP client configuration.

Setup and configuration

For the remote service, configure an MCP client with the Streamable HTTP endpoint; no installation or API key is required. The project supplies client configurations and usage instructions for Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Codex, and Gemini CLI.

A local stdio configuration can use:

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

Tools and capabilities

The analysis tools can return national FHA statistics, compare one lender with peers, list the top 100 FHA lenders by denial rate or volume, report state-level rates and small-loan splits, and summarize the Door Effect across lenders. Additional remote tools cover metro lender gaps, loan-size penalties, denial-reason shares, conditional lender surfaces, ready-made cohorts, and rule lookups such as FHA mortgage insurance and loan-limit tables.

check_claim_contract evaluates proposed wording against a figure's permitted use and can return a pass, qualifier requirement, or block, along with safe wording and a receipt when applicable. resolve_fha_annual_mip selects an FHA mortgage-insurance value from supplied loan conditions and requests missing conditions rather than guessing. Counterparty screening tools provide a free preview; a full brief requires a license purchased by a human.

Limitations and notes

Every FHA figure is a historical aggregate from the 2025 record. The lender analysis covers the 100 largest FHA lenders by volume, and observed differences should not be treated as individual underwriting outcomes or lender recommendations. The data is attributed to FinanceRateCalc and licensed CC BY 4.0. The local package is narrower than the remote endpoint and uses its bundled snapshot rather than the remote service's rule, receipt, and claim-checking features.