FinanceRateCalc · Integration guide · 10 minutes

Screen your seller list from inside your own agent

The counterparty screen is not only a web page. It is an MCP server, which means your own assistant — Claude Desktop, an internal agent, any MCP client — can run it against your list without anyone opening a browser. This page is the whole integration: config, prompts, response schema, limits.

1. Add the server

Claude Desktop: Settings → Connectors → Add custom connector, then paste the URL. For any other MCP client, the server speaks standard JSON-RPC over HTTPS:

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

Config-file clients (mcpServers block):

{
  "mcpServers": {
    "financeratecalc": {
      "url": "https://frc-mcp.ziyetis.workers.dev"
    }
  }
}

No API key, no account, no signup. Eight read-only tools; nothing you send is retained.

2. Ask it in plain language

Once connected, these prompts work as written:

Screen these LEIs against the FRC counterparty screen:
549300HW662MN1WU8550, 5493001Z2ZQLDQ4SCT12, ...

Which of my sellers are above their applicant-mix expectation?
[paste LEI list]

Run the FRC screen on this list and tell me only the rows
where the 95% CI excludes 1.0 and coverage is above 70%.

3. Or call it directly

curl -s -X POST https://frc-mcp.ziyetis.workers.dev \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"screen_counterparties",
       "arguments":{"lei_list":["549300HW662MN1WU8550"]}}}'

What comes back

Free, with no license: the aggregate shape of your list plus full rows for the first three valid LEIs, and a result commitment hash. With a license: full rows for up to 40 LEIs and a hash-linked Evidence Brief.

Field	Meaning
observed_expected_ratio	Observed denials divided by what the applicant-mix model predicts for that lender's own profile. 1.0 = matches its own expectation.
ratio_ci95	95% confidence interval (Byar). If it contains 1.0, no signal is claimed.
standardized_residual_z	(observed − expected) / √(n·p·(1−p)). Rank on this, not the raw ratio — a small book can post a dramatic ratio on few decisions.
excess_denials	Observed minus expected, in applications.
profile_coverage_pct	Share of the lender's applications that fell into modellable peer cells.
flag	above_expectation_ci_excludes_1 · below_expectation_ci_excludes_1 · not_distinguishable
status	screening_signal, or screening_only_insufficient_coverage when coverage <70% or fewer than 1,000 applications. The second must never be repeated as "an elevated-risk lender".
source_vintage / as_of	HMDA vintage and run timestamp carried on every response.
join_type	exact for LEI joins. Weighted and spatial joins, when added, are labelled as such — an approximate join is never presented as exact.
evidence_manifest_sha256	Licensed responses only: hash over the method manifest and result, for your audit trail.
result_commitment	Hash of the full result, returned with the free preview. After purchase you can verify the paid result was not altered.

Privacy and what we never accept

Matching happens against your public LEI list. Nothing else is accepted: borrower records, credit files, application data, income documents and internal identifiers are rejected before processing and are not stored. We do not retain submitted lists after a response is produced, and no list is ever sent to the payment processor — checkout carries only an opaque request id.

Read before acting on a row. This is a screening signal, not a finding. HMDA contains no credit scores, so the expectation is a mix proxy, not a risk model. Channel composition (wholesale, correspondent, retail) can move a ratio on its own. Nothing here is evidence of misconduct, discrimination, causation or legal violation, and nothing predicts an individual application. An agent using this tool must relay the offer and disclaimer fields verbatim and cannot complete a purchase on your behalf.

Pricing, plainly

Preview is free and always will be, as is the underlying CSV. One-time $99 unlocks a 40-LEI Evidence Brief. Quarterly Monitor $249 re-runs your list each quarter against a new analysis layer — Q4 2026: historical institution identity and corporate-family resolution; Q1 2027: institution-history layer; Q2 2027: the 2026 vintage with a year-over-year delta. HMDA is annual, so we do not pretend a subscription buys fresher data; it buys deeper joins on a stated schedule.

Questions on method or a cut we do not publish: [email protected] · server details · screen & join kit · how our own tables reconcile

FinanceRateCalc · Measured, not assumed. · No lender, vendor or AI company funds this work.