Skip to main content Skip to footer
Sign Up for Rate Alerts!

Live data from 1,000+ US lenders

LoanGlass Mortgage MCP Server

LoanGlass gives AI assistants live mortgage rate data from more than 1,000 US lenders, over the Model Context Protocol (MCP).

Endpointhttps://mcp.loanglass.com/mcp
  • Streamable HTTP
  • No sign-in
  • 3 tools: mortgage_rates, mortgage_lenders, mortgage_costs
1

What It Is

Published sources give you an average. LoanGlass gives you the distribution: what the cheap end of the market charges, the median, how far apart they are, and where one specific quote sits among them — and then it names the cheapest lenders, with their NMLS number and a link to the page the rate was read from. Percentiles, lender counts and named lenders are not available from any published source.

Two more tools answer the rest of the question. Rates tell you whether a number is good; mortgage_lenders tells you whether a lender is good, from their position measured every day over about three months — which separates a lender who is reliably cheap from one running a promotion this week. And mortgage_costs tells you what the loan actually costs: it carries the live market price of a discount point, measured daily from the lenders who publish the same loan at several price points at once, which is what settles whether buying your rate down is worth it.

Endpointhttps://mcp.loanglass.com/mcp
TransportStreamable HTTP
AuthenticationNone
Tools3 — mortgage_rates, mortgage_lenders, mortgage_costs
Rate limit300 requests per minute per client address
HealthGET https://mcp.loanglass.com/health

Why There Is No Sign-In

An assistant that cannot reach a connector does not tell you. It answers from the open web instead, and for mortgages that usually means a published national average — the one number this service exists to improve on. Every sign-in step makes that more likely, so there are none. The same rate data is already public on loanglass.com.

2

Connecting

Claude (Web Or Desktop)

Add a custom connector and paste https://mcp.loanglass.com/mcp. No authentication.

ChatGPT

Turn on developer mode, create a connector with the URL above, and set authentication to none.

Claude Code

claude mcp add --transport http loanglass https://mcp.loanglass.com/mcp

Other Clients

Clients that accept a configuration file:

{ "mcpServers": { "loanglass": { "type": "http", "url": "https://mcp.loanglass.com/mcp" } } }

Clients that only speak stdio can bridge to it with npx mcp-remote https://mcp.loanglass.com/mcp.

Calling It Directly

curl https://mcp.loanglass.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"mortgage_rates","arguments":{"state":"Ohio","myRate":6.75}}}'

The reply arrives as a stream with one data: line. tools/list returns the machine-readable schema and is the authoritative version of this page.

3

What To Ask

You talk to your assistant; it decides when to call LoanGlass.

  • “I was quoted 6.75% in Ohio — is that any good?”
  • “Who has the cheapest 30-year rates in Texas right now?”
  • “What are mortgage rates right now?”
  • “Is 5.99% competitive on a 15-year in Florida?”
  • “Am I being overcharged? I’m paying one point.”
  • “How much could I save by shopping around on a $600k loan?”
  • “Who should I call to beat the rate I’ve been offered?”
  • “Is Rocket Mortgage actually competitive, or just well advertised?”
  • “Is Navy Federal any good for a 15-year?”
  • “Are they cheap all the time, or just this week?”
  • “Should I pay a point to buy my rate down?”
  • “6.5% with 0.75 points, or 6.75% with nothing — which is better?”
  • “How long until paying points pays off?”
  • “I’m at 7.5% with 22 years left — should I refinance?”
  • “What do closing costs normally run in Ohio?”
4

Tool 1: mortgage_rates

One tool, two kinds of answer. Leave out myRate and you get a picture of the market. Include it and you also get a verdict on that specific quote.

Parameters

Every parameter is optional. {} returns the national 30-year fixed picture.

Name Type Default Notes
loanType string 30-Year Fixed One of the loan types below.
state string national Full name or two-letter code — Ohio or OH.
purchaseType string New Purchase New Purchase or Refinance.
propertyType string Single-Family Home The only covered value.
occupancy string Primary Residence The only covered value.
points string Zero Points Pass the borrower’s actual points if they pay any.
myRate number none A quoted rate as a percentage, e.g. 6.75.
loanAmount number 400000 Used to show differences in dollars per month.
termYears integer from loanType Leave unset. A 15-Year Fixed uses 15; ARMs use 30.
lenderCount integer 5 How many of the cheapest lenders to name, 0–25. Set 0 to suppress names; out-of-range values are clamped, not rejected.

Accepted Values

Matching ignores case, spacing and punctuation, so 30 year fixed and refinance both work.

Loan types

  • 30-Year Fixed
  • 20-Year Fixed
  • 15-Year Fixed
  • 10-Year Fixed
  • 30-Year Fixed Jumbo
  • 20-Year Fixed Jumbo
  • 15-Year Fixed Jumbo
  • 30-Year Fixed VA
  • 30-Year Fixed FHA
  • 10/1-Year ARM
  • 10/6-Month ARM
  • 7/1-Year ARM
  • 7/6-Month ARM
  • 5/1-Year ARM
  • 5/6-Month ARM
  • 3/1-Year ARM

Points

  • Zero Points
  • 0-1% Points
  • 1-2% Points
  • 2-3% Points
  • 3%+ Points
  • Any Points

Any Points mixes rates bought down with points and rates at par, and the response warns about it.

Recognised But Not Covered

These are real products that too few lenders publish to describe a market.

Parameter Not covered Answer instead on
purchaseType Cash-Out Refinance, Construction Refinance
propertyType Condominium, Multi-Family Home, Manufactured Home, Land Single-Family Home
occupancy Secondary Residence, Investment Property Primary Residence
Coverage is deepest for fixed-rate, new-purchase, single-family, primary-residence loans. Small states can fall below the reporting floors for less common loan types.
5

Responses: mortgage_rates

A Full Answer

Field Meaning
found true
as_of Date the underlying rates were observed.
basis Every assumption echoed back, including the observation window.
coverage lenders_reporting, rates_observed, confidence.
distribution p10_pct, median_pct, p90_pct, spread_bps, spread_monthly_usd and observed_range.
your_rate Only when myRate is supplied. See below.
cheapest_lenders The cheapest lenders by name. See below. Absent when lenderCount is 0 or nothing matched.
summary The whole answer in plain English, disclaimer included.
disclaimer Standing text.

your_rate contains percentile, lenders_cheaper, bps_vs_median, bps_vs_cheapest, monthly_vs_p10_usd, a plain-English verdict, and three fields about rounding: lenders_at_this_price_point, percentile_if_one_notch_higher and clustering_note.

Named Lenders

cheapest_lenders is the part that turns a percentile into a phone call. Each entry carries rank, lender, nmls, rate_pct, apr_pct, points_pct, monthly_principal_interest_usd, observed_on, and source_url — a link to the lender’s own page as we read it.

Ranking is by price alone. LoanGlass sells promoted placement on its own website; that placement is never applied here. Rows ascend by rate, and the first row sits at or below p10_pct in the same response, which a promoted lender pushed to the top could not do.

One row per lender, at their cheapest offer within the filter you asked for. Note that observed_range.lowest_pct describes the population the percentiles are drawn from, which is filtered slightly differently, so it can sit below the first row - it is not a check on the ranking.

Other Answers

Fewer than 10 lenders

Too Few Lenders To Describe A Market

Fewer than 10 lenders: sufficient_for_statistics: false and observed_rates_pct in place of distribution. The summary tells the assistant to widen the query rather than present a range.

found: false

No Lender Publishes The Combination

found: false with a suggestion. This is an answer, not an error.

invalid_argument

A Value That Isn’t Valid

No query is run. error: "invalid_argument" and a problems list, one entry per bad parameter:

  • reason: "unrecognised" — a naming problem. Includes did_you_mean and valid_values.
  • reason: "not_yet_covered" — a real product with too little data. Includes advice and covered_values. Call again on the covered value and tell the user which basis was used.
temporarily_unavailable

Temporarily Unavailable

error: "temporarily_unavailable" with retry_suggested: true. This covers both outages and planned maintenance. The summary explicitly tells the assistant not to substitute a figure from another source.

HTTP 429

Rate Limited

HTTP 429.

6

Tool 2: mortgage_lenders

mortgage_rates answers “is this number any good”. This answers “is this lender any good” — a different question with a different source.

The distinction that matters is cheap today versus reliably cheap. Any comparison site can rank lenders by this morning’s advertised rate, and a lender running a one-week promotion wins that list. What is not published anywhere is the same lender’s position measured every day over about three months: how often they actually beat the typical lender, where they sat on the daily leaderboard, and what their worst day looked like.

Parameters

Name Type Default Notes
lenderName string — Required. The name as the user said it. Partial names are matched loosely, so pass what they typed rather than guessing a legal entity name. Several matches returns a candidate list instead of a guess.
loanType string 30-Year Fixed Same vocabulary as mortgage_rates. A lender can be competitive on one product and not another.
purchaseType string New Purchase New Purchase or Refinance.

A Full Answer

Field Meaning
lender name, nmls, firm_type, regulator, city, state, website. regulator states FDIC or NCUA insurance with the certificate or charter number — the part borrowers mean when they ask whether a lender is legitimate.
basis window_days, days_ranked, window_start, window_end and peers_compared_daily. The window is reported, never chosen: position is recomputed daily against the lenders actually advertising that product, then aggregated.
track_record The measurement itself. See below.
confidence days_ranked and sufficient. Below 20 ranked days no verdict is offered, because a lender can look consistently cheap across a fortnight by luck.
data_quality Present only when a statistic was withheld. See Validation below.
verdict One phrase, e.g. “consistently among the cheapest lenders in the market, with steady pricing”. null when the sample is too thin.
summary The answer in a sentence or two, leading with the daily rank over the window.
disclaimer Standing text. Advertised rates, and a past pricing record is not a prediction.

The Track Record

Field Meaning
average_daily_rank Mean position on the daily leaderboard. Rank 1 is cheapest. Read it against peers_compared_daily.
best_daily_rank / worst_daily_rank The best and worst position reached on any single day.
pct_days_single_cheapest How often this lender was the outright cheapest.
pct_days_in_cheapest_quarter How often they sat in the cheapest quarter of lenders.
pct_days_cheaper_than_typical How often they beat the daily benchmark.
median_bps_vs_typical / avg_bps_vs_typical Distance from the typical lender in basis points. Negative is cheaper.
pricing_steadiness_bps Standard deviation of that distance. Lower is steadier pricing — this is what separates a reliably cheap lender from a lucky one.
biggest_swing_bps The widest gap between their best and worst day.
worst_day_pct_of_lenders_beaten On their worst day, the share of lenders they still beat. A floor rather than an average.
est_monthly_vs_typical_usd_400k The difference as dollars per month on a $400,000 loan. Positive means the lender saves money.

Validation, And Why A Field May Be Missing

Every statistic is checked against what a mortgage market can actually produce before it is returned. A spread beyond ±300 basis points, day-to-day variation beyond 150, or a monthly difference beyond $900 is treated as a data error, not a finding. The benchmark-relative percentage is also cross-checked against what the average rank implies: a lender cannot average rank 564 of 790 and also have beaten the typical lender on 46% of days.

A field that fails any of those checks comes back null and is listed in data_quality.withheld with the value that was rejected. Nothing is returned with a caveat attached, because a caveat next to a number does not stop the number being quoted. The rank statistics are order statistics and are unaffected, which is why the answer is built on them.

Other Answers

found: false

No Such Lender

The name matches nothing in the rate set. That means their advertised rates are not observed — it is not a judgement about the lender and does not imply they are small or new. A suggestion proposes a shorter form of the name.

ambiguous: true

Several Lenders Match

“First” matches dozens. Up to eight candidates come back by name so the assistant can ask which one is meant. An exact name match always wins outright rather than being offered as a choice.

found: false

No Record For That Product

The lender is in the rate set but has no measured record for that loan type and purchase type. Their identity details are still returned. Lenders only appear here for products they advertise consistently enough to rank.

invalid_argument

An Unrecognised Value

An unknown loanType or purchaseType is rejected with the valid list, rather than silently matching nothing and reading as a lender with no history.

Example

Asking for {"lenderName": "Navy Federal", "loanType": "15-Year Fixed"}:

{
  "lender": {
    "name": "Navy Federal Credit Union",
    "nmls": "399807",
    "firm_type": "Credit Union",
    "regulator": "NCUA-insured credit union (charter 5536)",
    "city": "Vienna", "state": "VA",
    "website": "https://www.navyfederal.org"
  },
  "basis": {
    "loan_type": "15-Year Fixed", "purchase_type": "New Purchase",
    "window_days": 92, "days_ranked": 37,
    "window_start": "2026-07-31", "window_end": "2026-10-30",
    "peers_compared_daily": 758
  },
  "track_record": {
    "average_daily_rank": 115.6,
    "pct_days_single_cheapest": 0.0,
    "pct_days_cheaper_than_typical": null,
    "median_bps_vs_typical": -54.37,
    "est_monthly_vs_typical_usd_400k": 119
  },
  "data_quality": {
    "withheld": ["pct_days_cheaper_than_typical (100%, against 85% implied by an average rank of 116 of 758)"],
    "instruction": "Do not estimate the withheld figures from the ones that remain."
  },
  "verdict": "consistently among the cheapest lenders in the market",
  "summary": "On 15-Year Fixed, Navy Federal Credit Union averaged rank 116 of about 758 lenders over the last 3 months. That is about $119/month cheaper than the typical lender on a $400k loan."
}
Rankings here are organic, exactly as in mortgage_rates. LoanGlass sells promoted placement on its own website; that placement is never applied to anything this server returns.
7

Tool 3: mortgage_costs

The first two tools answer whether a rate is good and whether a lender is good. This one answers what the thing actually costs — the question that decides whether you pay points, take the credit, pick quote A over quote B, or refinance at all.

The part that exists nowhere else is the price of a discount point. A point costs 1% of the loan; what it buys is a market price that moves every day. We measure it from the few hundred lenders that publish the same loan at several price points at once, by fitting rate against points for each of them. The familiar rule of thumb — that a point buys a quarter of a percent — is wrong most days, and nothing published anywhere will tell you which way.

Higher is better

The figure is returned as bps_of_rate_per_point: basis points of rate that one point buys. When it falls, a point buys less and buying down gets more expensive. Our own market dashboard shows the same thing upside down, as points needed per quarter percent, so read the field name before the number.

Parameters

Every one is optional. With none at all you get today’s price of a point.

Name Type Default Notes
loanAmount number 400000 The loan, not the purchase price and not the payment.
rate number — The quote’s rate as a percent, e.g. 6.75.
points number 0 Percent of the loan. 1 is one point; negative is a lender credit.
compareRate number — A second quote, to settle a head-to-head.
comparePoints number 0 Points on that second quote.
termYears number 30 For a refinance this is the new term, which is what makes a re-extension visible.
yearsBeforeYouLeave number 7 How long the loan is kept. The single biggest driver of every verdict here.
lenderName string — Check whether one lender prices its own points better or worse than the market.
state string — Adds what borrowers there actually paid in closing costs.
currentRate number — Refinance: the rate being paid now. Use rate for the new one.
currentBalance number — Refinance: what is still owed. In a refinance this becomes the loan amount.
currentYearsRemaining number termYears Refinance: years left. Supply it — the default hides the cost of re-extending.
refinanceClosingCosts number 0 Refinance: dollars, excluding points. Zero flatters the refinance, and the response says so.

What comes back

An answering list names the sections that were filled, so you can tell a question that was answered from one that was not asked.

Section When What it holds
point_price Always Today’s price of a point, the middle half of measured lenders, the lender count, and a ten-day trend with what it implies about where the market thinks rates are going.
quote rate given Points cost, monthly payment, and the par-equivalent rate — what the quote amounts to at zero points. That is the number to compare offers on.
buydown points above zero Whether the points are worth paying: break-even, the cost each way over your horizon, and a verdict.
comparison compareRate given The head-to-head, decided on cost over the holding period rather than on the headline rates.
lender_point_price lenderName given That lender’s own measured point price against the market’s.
refinance currentRate and currentBalance Break-even, cost both ways, and a term-extension warning.
typical_closing_costs state given HMDA closing costs for that state, with what each average does and does not cover.

Break-even, done properly

Almost every published break-even divides the cost of the points by the monthly saving. That ignores the fact that a lower rate also pays the loan down faster, so the usual figure comes out later than the truth. On $400,000 at 6.5% with 0.75 points against 6.75% with none, the familiar sum gives 45 months; counting the equity gives 36. Both are returned — break_even_months and simple_break_even_months — so a lender’s own number can be checked against the one it was probably calculated with.

The same horizon decides the answer. Those two quotes are not simply better or worse than each other: the one with points wins if the loan is kept seven years and loses if it is kept two. Ask the user how long they expect to stay before quoting either verdict.

Refinancing, and the term trap

Stretching 22 remaining years back out to 30 lowers the payment whatever the rate does, so a monthly “saving” can be pure term extension. Refinancing $320,000 from 7.5% with 22 years left to 6.5% over 30 years cuts the payment by $456 — and adds about $74,000 of interest over the life of the loan. Keep the 22-year term and the payment falls by $197 instead, which is worth $5,700 more over seven years.

So term_extension_months and a plain-language warning come back whenever the new term outruns the old one, and the verdict is always decided on cost rather than on the monthly figure.

Other answers

Situation What comes back
Too few lenders measured The point price is withheld and listed in data_quality.withheld, rather than returned with a caveat next to it.
Lender publishes one price a day found: false with an explanation. Most lenders cannot be measured, and that is normal rather than a red flag.
Half a refinance answered: false naming the missing arguments, instead of the section silently vanishing.
A rate of 675, or a payment passed as the loan invalid_argument listing each problem. A mistyped rate produces a confident, catastrophic answer, so these are refused rather than worked around.
Pricing data unreachable The arithmetic still runs. A two-quote comparison needs no data at all, so it survives an outage — and the response says explicitly not to fall back on the quarter-percent rule of thumb.
8

Reading The Numbers

  • Percentiles count lenders, not rows. Each lender counts once, at their best offer, so a lender publishing many rates cannot tilt the result.
  • The headline spread is the 10th to the 90th percentile. One lender running a teaser rate would otherwise double what shopping around appears to be worth. The raw lowest and highest rates are in observed_range, labelled as observed rather than achievable.
  • Lenders price in eighths, so many sit on exactly the same rate. A quote just under a crowded price point can move many percentile places for a tiny change. clustering_note says when that applies.
  • Confidence: with 30 or more lenders you get everything. From 10 to 29 there is no percentile and no verdict. Below 10 there is no median either.
  • Verdict bands (30+ lenders only): up to the 10th percentile excellent, up to the 30th good, up to the 55th reasonable, up to the 80th below average, above that poor.
  • Monthly payment figures use the term implied by the loan type. ARMs are modelled over 30 years, because “7/1” is how long the rate is fixed, not how long the loan runs.
Excellentto 10th
Goodto 30th
Reasonableto 55th
Below averageto 80th
Poorabove 80th

Example

Arguments {"state": "Ohio", "myRate": 6.75} — response, trimmed:

{
  "found": true,
  "as_of": "2026-11-13",
  "coverage": { "lenders_reporting": 74, "rates_observed": 88, "confidence": "high" },
  "distribution": {
    "p10_pct": 6.875, "median_pct": 7.234, "p90_pct": 7.625,
    "spread_bps": 75, "spread_monthly_usd": 203,
    "observed_range": { "lowest_pct": 6.250, "highest_pct": 8.000 }
  },
  "your_rate": {
    "rate_pct": 6.75, "percentile": 4, "lenders_cheaper": 3,
    "bps_vs_median": -48, "bps_vs_cheapest": 50, "monthly_vs_p10_usd": -33,
    "verdict": "excellent - among the cheapest available",
    "lenders_at_this_price_point": 6, "percentile_if_one_notch_higher": 12
  },
  "cheapest_lenders": [
    {
      "rank": 1, "lender": "Huntington Federal Savings Bank", "nmls": "412667",
      "rate_pct": 6.250, "apr_pct": 6.287, "points_pct": 0.000,
      "monthly_principal_interest_usd": 2463, "observed_on": "2026-11-13",
      "source_url": "https://loanglass.com/go/1-ifkzp-1fykuvn07f64v-o-s-1-aqc5.p8mqSOkSU-7puzVKrT6feA"
    }
  ]
}
9

About The Data

Source Rates lenders advertise on their own websites, collected daily by around 300 collectors.
Freshness Each lender’s most recent rate within the last five days. as_of is the date observed.
What it is not Advertised rates are indicative, not offers, and not financial advice. A borrower’s actual rate depends on credit, loan-to-value, points and property.
No paid placement Nothing sponsored influences what this returns.
10

Privacy

LoanGlass never receives your conversation — only the arguments the assistant chose to send. Each call records those arguments, the assistant’s name and version, a session id, timing and the outcome. None of that identifies a person.
11

Service Status

GET /health returns {"status":"ok","answering":true}. answering: false means the service has been switched off deliberately, and every question receives a stated maintenance response rather than a dead connection.

Rates are advertised by lenders on their own websites and observed by LoanGlass on the date shown. They are indicative, not offers.

Disclaimer

LoanGlass is not a lender. LoanGlass (formerly mortgage-rates.ai) is an independent site that helps borrowers compare mortgage rates. We are not a lender, mortgage broker or financial advisor, and we are not registered with the Nationwide Multistate Licensing System (NMLS). Nothing on this site is an offer to lend.

Where the rates come from

Every day we collect the rates that more than 1000 lenders publish on their own websites, and we show them as published. Averages, ranges, charts and forecasts are our own calculations, and forecasts may be wrong. Lender records, reviews and market data come from public sources such as the NMLS, NCUA, OCC, HMDA filings, the U.S. Treasury, Freddie Mac and Bankrate. We work to keep it all accurate, but it may contain errors.

Your rate may differ

The rate a lender offers you depends on your credit, income, down payment, loan and property, and you may not qualify for every rate shown. Payments shown with rates are principal and interest only; taxes and insurance are extra. Calculator results are estimates. Ask lenders for an official Loan Estimate before you decide.

Promoted listings

Lenders are included for free, but some pay for promotion. Promoted lenders are marked "Promoted" and appear above other lenders in rate results, whatever their rate. Paid placement can also decide the order between lenders that show the same rate. Paying never changes the rates a lender publishes.

Questions or corrections?

To report an error or have your institution's rates included, email content@loanglass.com.

All logos, trademarks and brand names belong to their owners.

We use cookies and similar technologies to improve your experience, analyze site usage, and deliver personalized content. By clicking "Accept", you agree to the storing of cookies on your device in accordance with our Privacy Policy. You can manage your cookie preferences at any time.