RateAPI — live US mortgage, auto, HELOC, personal & deposit rates

Add recipients (email + optional name) to a campaign. Up to 50 per request, 200 per campaign. Duplicate addresses are ignored.

Compare projected cost of renting vs buying a home over a holding period, anchored to LIVE market mortgage rates. Given home price, down payment, monthly rent, and a holding period, returns year-by-year cumulative cost of each path (rent vs PITI), equity built, net cost of owning after sale, and the breakeven year where buying becomes cheaper than renting. If no rate is supplied, the current best market rate for the state is used. Use for 'should I rent or buy?' questions.

Frame a buy/refi-now vs wait recommendation. Combines recent rate movement (direction + magnitude over 1w/2w/4w) with break-even sensitivity on the borrower's loan to return a forward 'lock now' vs 'wait' lean, the bps swing that would change the call, and the cost of waiting. Use for 'should I lock now or wait for rates to drop?' questions. Distinct from get_rate_delta (raw historical deltas only) — this adds a forward recommendation. Not a rate forecast.

Lock-window-aware lock-vs-float-down recommendation for a deal already in the pipeline. Given a known lock-expiration date, the currently locked rate, and the float-down cost, returns a defensible hold-lock vs exercise-float-down call a loan officer can document — combining recent market movement (1w/2w/4w) with payment and break-even math on this loan. Builds on the same recent-movement signal as assess_buy_timing but framed around the lock window/expiration. Not a rate forecast.

Score an array of inline financing scenarios in one stateless call; returns ranked decisions and top offers per item — no persisted cohorts, no per-borrower HTTP loops.

Build a year-by-year amortization / payoff schedule for a loan: total lifetime interest, principal/interest split per period, and a what-if extra-payment accelerator showing months saved and interest saved. Resolves the market rate from real credit-union offers when no rate is provided.

Run compliance/suitability guardrails over a financing scenario and an optional recommendation. Returns flags (red/yellow with code + message) for things like DTI strain, long auto terms with negative-equity risk, possible balloon payments, and stale market data, plus generic disclaimers and an overall verdict (ok/caution/stop). Use this before presenting a rate recommendation to a borrower.

LENDER-side 'is my offer competitive' check — input your OWN published rate for a state+product, get the market percentile, bps vs the state median, and a leading/competitive/overpriced verdict. This is for a loan officer or credit union benchmarking their published pricing against the live market distribution. Do NOT use this for a consumer's quote: if the user is a BORROWER asking whether a rate they were offered is fair, use rate_sanity_check instead.

`relative_memberships` carries confirmed current/deceased relative membership scoped to one exact target credit union; one institution is never reused for another. Check whether a user can JOIN specific credit unions (field-of-membership eligibility). Batch-friendly: pass 1-50 credit_union_ids (from get_financing_decision offers, list_credit_unions, or get_credit_union) plus whatever is known about the user (home, worksite, and payroll geography are independent facts; employer, occupation, military_status, school plus school_relationship, and worship are also supported). City facts require their same-role state or a recognized same-role ZIP that uniquely identifies one state, and are accepted only when the closed Census gazetteer resolves them. Returns one row for every requested credit union. Positive outcomes include open_to_anyone, likely_eligible, eligible_via_employer, and eligible_via_military. A school name alone is not a positive signal: school_relationship is required to affirm a typed student, alumni, or school-employee path. Only a graph carrying exact active-publication lineage governs the top compatibility verdict; a publication-less shadow graph remains diagnostic and cannot replace the flat result. A non-match is returned as unknown, with `unknown_reason: fom_completeness_unaffirmed` when the published field-of-membership corpus is not attested complete. The legacy `geo_mismatch` enum value is reserved for wire compatibility and is not emitted by this MCP boundary without a future typed closed-world attestation. Verdicts are guidance based on each credit union's published membership criteria; the final membership determination is always the institution's. Non-matches carry `missing_facts` for the named institutions: the exact question to ask next, the `fact_field` an answer populates, and the institution's own words. A check that returns unknown for all of them and a question for each is a SUCCESSFUL call, not a failure — ask the questions, merge the…

Compare baseline and candidate confirmed person facts across the same 1-19 explicit products. Both catalogs use one pinned immutable eligibility graph snapshot; rate rows retain independent evidence and as_of. Returns access, membership-path, and like-for-like published-price basis-point deltas: APY for deposits, APR for lending when both sides publish APR, otherwise nominal rate only when both sides publish nominal rate. It never mixes APR and nominal rate. Never describe missing access as denial or infer savings, underwriting, approval, or final terms.

Compare the best real rates across MULTIPLE lending products in a single call — e.g. show a buyer mortgage vs HELOC, or new vs used auto. Runs the decision engine for each product and returns a ranked, true-cost (APR + fees) table. Use this instead of calling get_financing_decision repeatedly. Each product returns its top offer with rate, APR, and monthly payment.

Compare one baseline person-fact panel with one candidate panel against the same active eligibility snapshot and exact product request. Use this for explicit what-if questions such as adding a confirmed employer or association. Newly reachable means current positive evidence now proves a path; no_longer_reachable or absence never proves ineligibility. Supply only facts the person confirmed, preferably canonical *_org_ids from resolve_eligibility_entities. Person facts are forwarded transiently and are not stored by this tool.

Apples-to-apples multi-lender ranking by TOTAL cost of a loan (upfront origination/lender fees + lifetime interest over the holding period), not just the headline rate. Re-ranks real credit union offers and flags when the lowest-rate lender is not the cheapest overall.

Subscribe a monitor to a SPECIFIC named credit union's published rate and alert via webhook when that CU changes its rate (vs the aggregate market-best decision). Returns the monitor with a webhook_signing_secret (shown once) and a baseline snapshot of the CU's current rate.

Create a branded rate-email campaign: on a cadence, email the campaign's recipient list the latest rates for one or more states (or a named region preset) under the org's branding, with ONE OR MORE products (1-5, no duplicates). Each product is one of: {category:'mortgage', subtype:'30-year-fixed'|'15-year-fixed'}, {category:'auto', condition:'new'|'used'|'any', term_months?:36|48|60|72|84} (auto = cars only), or {category:'rv'|'boat'|'motorcycle', condition:'new'|'used'|'any'}. Defaults to a single 30-year-fixed mortgage if products is omitted. Created 'paused'; add recipients, then activate via update_email_campaign (activation requires branding display_name + business_address). Requires a Pro plan. Provide exactly one of state_code, state_codes, or region.

Create a new decision monitor that watches for rate changes and sends webhook notifications when conditions are met. Returns the monitor with a webhook_signing_secret for verifying webhook payloads.

Permanently delete a saved cohort by id. This action cannot be undone.

Delete a scheduled digest by id. Stops the recurring email.

Delete a branded rate-email campaign and its recipients. This cannot be undone.

Permanently delete a monitor. This action cannot be undone.

Delete a saved scenario by id.

Send a test webhook for a monitor using a simulated rate change. This actually delivers an HTTP request to the monitor's configured webhook URL. IMPORTANT: a test delivery is NOT signed with the monitor's signing secret — only a hash of that secret is stored, so test deliveries use a placeholder and will FAIL signature verification. Use this to test payload handling, not signature verification. Only use when the user explicitly requests a test webhook delivery.

Borrower-facing home affordability estimate: given gross income, existing monthly debts, down payment, and front-end/back-end DTI caps, compute the max monthly payment, max loan amount, and max home price. Surfaces the max_loan annuity engine. If no rate is supplied, the current best market rate for the state is used.

Total monthly housing cost (PITI): principal+interest plus property tax, homeowners insurance, PMI, and HOA, so affordability reflects real monthly outlay. If no rate is supplied, the current best market rate for the state is used.

Evaluate a caller-supplied panel of 1-10 named fact profiles for one exact product against one serving snapshot, deduplicating reachable institutions and exact offers. This is descriptive only for the supplied panel: never call it a population estimate, conversion rate, approval forecast, or negative eligibility result. Use neutral profile labels instead of names, emails, account numbers, or other direct identifiers. Profiles are forwarded transiently and are not stored by this tool; unknown remains unknown.

Explain WHY each top lender ranked where it did for a financing decision. Takes a full decision scenario (same shape as get_financing_decision) and returns a structured per-offer cost breakdown (rate, points, fees, lifetime interest), scoring signals, how much each offer beats the next by, and a plain-language summary of the ranking weights. Use this when a user asks why one lender beat another.

Turn a person's own words about themselves into the structured facts the eligibility tools accept. Pass their sentence verbatim ("I'm a nurse at a Tulsa hospital, I live in Broken Arrow, my late father was a TTCU member") and this returns each fact it could read, with the exact words it read it from, a confidence, and a needs_confirmation flag. THIS DOES NOT DECIDE ELIGIBILITY and never returns a credit union: it only fills in fields. Show the extracted facts to the user and get confirmation on every fact with needs_confirmation=true (family relationships, relatives' employers and associations, applicant kind, and asserted qualifiers are always flagged, because a wrong reading of those changes the answer). Then pass `person_search_body` — plus whatever the user confirmed — to find_eligible_credit_unions, which is what actually decides. If extraction is unavailable you get facts: [] and no eligibility signal at all; ask the user for fields directly instead. `confidence` is the extraction model's own uncalibrated self-report and gates only autofill — quote `evidence.quote` to the user, not the number. This call costs 3 API units.

Use when someone asks for "credit unions near me," the closest credit union, local credit-union branches, or nearby alternatives to a bank. Given a 5-digit ZIP, returns credit unions ordered by the estimated straight-line distance to each institution's closest NCUA-reported branch. Includes approximate miles, branch address, and profile identifiers for checking rates, services, and membership. Distances use ZIP centers and are not driving distance.

`relative_memberships` carries confirmed current/deceased relative membership scoped to one exact target credit union; one institution is never reused for another. `next_questions.unlocks_cu_count` counts membership decisions an answer may clarify, never guaranteed products. Discovery tool: find candidate credit unions a person may be able to JOIN, from their own facts rather than from a list of institutions. Pass any combination of home, worksite, or payroll geography (these are independent), occupation, employer, school (+school_relationship), associations, military_status, and worship; at least one discriminating fact is required, or pass willing_to_join_association=true as the sole discriminator to search documented joinable-association paths. City facts require their same-role state or a recognized same-role ZIP that uniquely identifies one state, and are accepted only when the closed Census gazetteer resolves them. This evaluates a bounded, prioritized candidate set, not the entire credit-union fleet: read `candidate_limit`, `candidate_set_truncated`, and `candidate_sources`, and never interpret an absent institution as unavailable or ineligible. Returns four buckets within that candidate set — eligible (a verified path already applies), conditionally_eligible (one more step, e.g. joining an association, with the cost and join URL when known), possibly_eligible (a path may apply but a requirement could not be confirmed), and unknown (we could not decide — either we hold no rules for the institution or the supplied facts cannot decide the rules we hold). Every non-positive entry carries a machine-readable `unknown_reason`: `geo_inputs_missing` / `org_inputs_missing` / `inputs_missing` mean ONE MORE FACT FROM THE USER would resolve it, so ask for it and call again; `no_rules` means we hold nothing for that institution and asking will not help; `incomplete_fom`, `fom_completeness_unaffirmed`, `low_confidence_rules` and `unmodeled_condition` mean we hold rules bu…