Everything we have shipped.

The hosted MCP endpoint at https://api.onegoodarea.com/mcp now supports OAuth sign-in. In Claude, add it as a custom connector and sign in with your OneGoodArea account; in Claude Code, add it with claude mcp add --transport http and authenticate from /mcp. Approving creates an API key named after the app in your account, metered like any other, and revoking it from your dashboard disconnects the app. Header keys keep working for clients that cannot sign in.

The twelve MCP tools are now served by the API itself at https://api.onegoodarea.com/mcp over Streamable HTTP, so a client reaches them with nothing to install. Every tool call runs through the same API routes as a direct call, under the same key, plan and allowance. The npm package and the MCP Registry listing continue for clients that spawn a local server.

The public oga-mcp repository now carries a Claude plugin: the hosted connector plus four skills that teach Claude how to research an area for a lender, compare sites for retail, read confidence and observation dates, and watch a portfolio for change. Each skill is written from the methodology and the tool definitions. Until the Claude directory lists it, claude --plugin-dir runs it locally.

MCP server 1.0.13 adds resolve_place: a postcode, place name, official code or place id becomes the permanent place it names, with how it resolved and whether a postcode is live or retired. Every tool now carries a title and a read-only hint, and its description names the coverage as England, Wales and Scotland.

The free sandbox moves from 200 to 500 calls a month, with a ceiling of 100 a day, and from 10 to 100 natural-language questions. Sandbox questions now answer on a paid model first, so a first question after a quiet spell no longer waits on a free host.

Every plan with API access now includes the MCP server, from the free sandbox to the platform licences. Nothing extra to buy.

New API keys no longer feed natural-language questions and explained scores into OneGoodArea's training logs unless their owner switches capture on. Existing keys that never changed the old default are switched off too; the Training switch in the console opts a key back in.

Each nation's deprivation index is now read by domain: income, employment, health and education in England, Wales and Scotland, crime in England and Scotland, and each nation's own further domains. A domain is a decile (1 is the most deprived) with a nation percentile, and a new optional detail field carries the published rank and, where the publisher gives one, the score. Scotland publishes a rank but no score.

Deprivation deciles and ranks are relative within each nation, so ranking areas by any deprivation signal, including deprivation.imd_decile and deprivation.imd_rank, now needs country or lad. Without one, /v1/areas returns 400 and /v1/query returns 422 with the message: Deprivation deciles and ranks are relative within each nation. Pass country=England, Wales or Scotland, or a lad.

Welsh deprivation is now read from the Welsh Index of Multiple Deprivation 2025, published on 2021 neighbourhoods, in place of WIMD 2019. Welsh deciles and ranks change with the new edition, so a neighbourhood can sit in a different decile than before.

Flood areas within 3km are counted the way the Environment Agency's own lookup counts them, from the list and outlines of 5 October 2026. The previous file under-counted: the average English neighbourhood has about 8 flood areas within 3km, not 2.

The Environment score now ranks an area's flood exposure among neighbourhoods in the same nation, the least exposed highest, and the reasoning states the rank. Before, a fixed amount came off per flood area, so nearly half of English neighbourhoods scored the maximum and the score could not tell them apart.

Two new signals count the Ofsted-graded schools within 1km of a neighbourhood's centre and the share rated Good or Outstanding, beside the existing 1.5km figures. They follow the same confidence rules, and they are not used when matching similar places, so peer sets did not move.

GET /v1/area takes grain=neighbourhood, group or district. At group and district level, crime counts are summed across the neighbourhoods inside, the rate per 1,000 residents is recomputed over the full twelve months from the population of the places counted, and national ranks are left empty rather than estimated.

The public pages for 31 cities now read the live area profile, its place and one explained score, refreshed daily, on an engineering account of their own. Each page is named after the neighbourhood it describes, carries the methodology version and the dates behind every value, and publishes Place and Dataset structured data.

A postcode Royal Mail has retired now resolves to the neighbourhood it was in, and a place id can be passed in area= directly, as /v1/score and /v1/peers already allowed. The response says whether the postcode is live or retired and the month it was retired. Before, a retired postcode returned 404 and a place id 400.

Every amenity count now comes from a dated OpenStreetMap extract through a generator we keep under version control and check before anything is saved. Parks, schools and other places drawn as outlines count when any part is within reach, matching the live lookup.

Gardens and swimming pools now count towards parks and leisure only when they have a name, so a neighbourhood where back gardens or garden pools are mapped no longer looks greener than it is.

A newly issued postcode used to be compared with suburban neighbourhoods whatever kind of place it was in. It now takes its neighbourhood's official area type. Where none is known, the area type is left empty and transport, schools and amenities are ranked against all of Great Britain, with the reason given on each.

Each scheduled source load is checked for completeness, coverage of every council, the months it should hold and the columns it needs. A load that fails is held back, the previous values stay live, and the reason is recorded.

In England, a score made without Ofsted data, for example with a key whose bundle leaves schools out, used to read as no graded school nearby and pull the score down. The Schools dimension now scores at the neutral midpoint and says why.

Change reports and feed entries now say whether the source data moved a value or a OneGoodArea methodology release did, and name the release. Webhooks send source-data changes by default, including for existing subscriptions; create a subscription that asks for methodology changes to receive those too.

The help pages and dashboard said a failed delivery is retried. It is not: a delivery that does not get a 2xx within 5 seconds is recorded as failed. Alert on failures at your endpoint; the change report and feed still list every change.

When HM Land Registry was slow or failed, the sales list for an area answered that there were no sales in the last 12 months. It now says the source did not answer, and a busy area's year-on-year price change is withheld when Land Registry returns only part of the earlier year.

A forecast of a count, a price or a rate is floored at zero; a short series could project a negative house price. Each projected month now carries its dates, and the documented example returns a forecast.

A value that is not available, such as schools in Wales, now has an empty period instead of a phrase like "Latest inspections"; the reason says why it is missing. A live source that did not answer is marked as unavailable.

A delivery that does not get a 2xx within 5 seconds is now sent twice more within about 30 minutes, with the same delivery id so your endpoint can ignore a repeat, and is marked exhausted after the third attempt. The dashboard shows the latest success and failure of each endpoint.

Bus stop and station counts are now dated by the day their NaPTAN file was taken, and England flood-area counts by the day the Environment Agency list was taken, instead of the day the data entered OneGoodArea. Amenity counts from OpenStreetMap and the deprivation indices keep their load date until their source states one; the deprivation indices now carry their publication dates.

Every value has a release date field, but it was always empty. It now carries the publisher's release date where the publisher states one, such as ONS house prices, and sorting by release date works. A source that states no release date, such as police.uk, leaves it empty instead of borrowing the data's own month.

The sales list for a postcode now answers in under a second instead of up to 45. It covers the postcode's neighbourhood and lists every sale, including repossessions, sales to companies and non-residential property.

With seven of the last twelve months missing from police.uk, Gloucestershire's 393 neighbourhoods averaged 35 recorded crimes per 1,000 residents against 98 nationally. Street-level crime is no longer served there; the figures explain why and show the Home Office council-area rate as context.

Median house prices, their change on the year before and the number of sales now come from ONS's year ending March 2026 release, published on 17 September 2026.

The OneGoodArea MCP server is listed in the official MCP Registry as com.onegoodarea/oga-mcp. Clients that install from the registry can add it by name, with your API key as the only setting. The config block on the MCP docs page still works everywhere else.

Each signal carries the period it describes as a start and end date, whether the source stated that period or it is the day the value first entered OneGoodArea, when it was loaded and, for a value fetched live, when it was fetched.

Ask for an area's signals newest first by the period they describe, by when they were loaded or by the publisher's release date. The MCP server supports the same ordering from version 1.0.7.

Ranked areas keep their order by value and now carry each value's dates, so a ranking shows how current every figure is.

History is now kept by date, so a value that changes always gets its own entry. Before, some changes to amenity counts were not recorded in history.

Rankings, similar places and insights filtered to a Scottish council area returned no areas. They now return every neighbourhood in it: 597 in the City of Edinburgh, for example.

Crime totals and rates now cover the twelve calendar months ending at the latest police.uk release for every neighbourhood, counting a quiet month as zero. Before, a quiet area's figure could reach back further than a year.

Greater Manchester Police has not published street-level crime since 2019, so the few records police.uk holds made its neighbourhoods look like the safest in the country in rankings and similar places. They are no longer ranked; the council-area crime rate is still shown.

When a neighbourhood had none of a crime type in the year, that type showed as not published. It now reads 0, for every type and every neighbourhood, and the fourteenth police.uk type, other crime, is counted so the types add up to the total.

Where street-level crime is not published, such as Greater Manchester and Scotland, the Crime part of the score now gives the same reason as the crime figures and says it was scored at the neutral midpoint.

Half of all ranked neighbourhoods have every nearby school rated Good or Outstanding. Because they all shared that top value, they were ranked at the 49th percentile. Tied values now share the middle of their group, the rule our scores already use: a perfect record now reads at the 74th percentile, and the least deprived decile at the 95th.

Similar places are matched on the corrected positions, so a few peer sets change once.

Crime trends over a year, six months and two years now cover exactly those calendar months everywhere, counting a month with no recorded crime as zero. Before, a quiet month could stretch a trend back further than it said, or leave an area with no trend at all.

The number of homes sold now covers the same year as the official ONS price and its change, so every property figure describes the same twelve months. Before, it counted this year's sales registered so far, which undercounted busy areas.

We have retired four neighbourhood property trends: two price changes built from as few as ten sales, which could contradict the official figure, and two sales-volume trends distorted by the months Land Registry is still registering. The official ONS change over the year is the price trend, for every area.

Crime types carry their official police.uk names on every surface, transport figures credit the national stop register, and every date and count reads the same way wherever you see it.

Where an area's price is the official ONS figure for its neighbourhood group, the confidence on the Property score now follows the sales behind that figure. Before, it could read low because it counted only a handful of recent sales nearby.

Every score explanation now names the basis of the price it uses and credits the right source, reads as plain sentences, and leaves out anything that is not known instead of showing a blank.

Every neighbourhood group in England and Wales, and every neighbourhood in Wales, now carries its official ONS name wherever a place is shown.

Scottish areas were sometimes looked up under a newer area code than the one their data is held under, so finding similar places, forecasts, change monitoring and some portfolio and area look-ups returned little or nothing. Every endpoint now finds an area the same way, from our own data, and Scottish areas are answered in full across all of them.

Before, two areas could be counted as similar when they were alike on only a few measures. A similar place must now match on most of them, and each area's set of similar places is refreshed in full every time. The measures that compare an area with places like it are recalculated on that basis.

Scottish and Welsh flood risk is measured for every neighbourhood from SEPA and Natural Resources Wales flood maps, refreshed every quarter, and answered instantly from our own data. Each flood figure now names the agency behind it and says what it measures, since the three agencies map flood risk differently. Real-time flood warnings are shown only when they were checked, and are clearly marked when they were not.

Some schools have no Ofsted grade yet, mostly new schools waiting for a first inspection. Every school within 1.5km is now counted, so a figure reads "9 of 12 schools carry a grade" rather than just 9, and an area whose schools are all waiting for a grade no longer looks as if it has none. Each figure gives the span of inspections behind it and the median age of the grades. Confidence is lower where a quarter or more of nearby schools are ungraded, or where most grades are more than six years old. Scores are unchanged.

Save a list of your own areas, such as a lending book or a store estate, and a score can compare an area against them instead of against similar places nationally. Postcodes and area codes are both accepted, and anything that cannot be matched is listed before the list is saved. Every score says which comparison it used, and a list too small to compare against fairly is refused rather than guessed.

Before, fixed benchmarks gave about half of city neighbourhoods the top score for schools and amenities, so those scores could not tell them apart. Each area is now ranked against every neighbourhood of its kind in Great Britain, and the comparison is rebuilt with every data refresh, so it keeps pace with the data. Typical areas now score near the middle, and scores near the top mean an area really is among the best served.

A school's grade now comes from its most recent Ofsted judgement: a report card, an inspection grade, or a short inspection that confirms the grade. Schools with a grade rise from under 10,000 to nearly 20,000, and the typical neighbourhood has 7 graded schools within 1.5km instead of 3. Schools and amenities scores rise where schools were missing.

Every school in the schools-in-range list names the Ofsted judgement behind its grade, and schools not yet inspected are listed too.

Each station past the first used to take points away without limit, so central Manchester, with 18 stations within 2km, scored lower on stations than a town with five. Stations now add points with diminishing returns up to five and level off after. About 3,200 city neighbourhoods score higher; areas with five stations or fewer are unchanged.

Every area is classed from the Office for National Statistics and Scottish Government rural-urban classifications of its postcode, using the newest boundary between town and country. Before, the classification could not be read and every area was scored against the suburban benchmark; city and rural scores now reflect the kind of place they are.

Every area response names the official rural-urban class it was benchmarked on, so a score can be traced to the kind of place it assumed.

Some transport authorities publish their bus stops with Ordnance Survey grid references only, and those stops were left out of the counts: Norfolk, Cambridgeshire, Peterborough, Cumbria, Shropshire, Angus and the Glasgow and Strathclyde councils read zero or close to zero bus stops. Every stop in the national register is now counted, so transport signals and scores there reflect the stops that exist.

Ask what has changed among the places in a portfolio since any date. Every change shows the value before and after, its confidence, where it sits nationally, and when it was recorded. Where an official boundary change has retired a place you hold, the feed shows where it went. Deliveries by webhook after each data update.

Send a portfolio of postcodes or official area codes, up to twenty thousand rows, and get every row back with the place it belongs to and that area's signals. Rows that cannot be matched come back with the reason, so nothing is silently dropped. Large sets run in the background and can be read in pages.

Every neighbourhood, district, region and nation in Great Britain now has a OneGoodArea place id that stays the same when official statistical boundaries are redrawn. Postcodes, official codes and place names all resolve to it, and codes from earlier census geographies still work. Retired postcodes resolve to the place they were in.

The inspected schools behind an area's schools figures can now be listed by name, nearest first, with their phase, grade and inspection date. The names and the numbers come from the same list, so they always agree.

Confidence on every schools figure now rises with the number of inspected schools nearby, and the reason on each value says how many stood behind it.

A share based on one or two schools is still shown, but no longer given a national percentile. No counts or shares change.

The schools dimension of a score now carries the same confidence as the schools signals, so a single school can no longer make a score look certain.

Every area read now resolves on OneGoodArea's own place registry and returns the area's permanent place id beside its codes, so a portfolio, a feed and an area response all name the same place. Retired postcodes resolve to the place they were in, with the month they were retired, on every area endpoint.

Every neighbourhood in England and Wales returns its Census 2021 resident count alongside the crime rate. No values change.

Recorded crimes per 1,000 residents over the latest twelve months, for every neighbourhood in England and Wales, with its own national percentile. The raw count remains.

The rate is shown only when a full year of data is present; a shorter period is never scaled up. Where street-level crime is not published, the rate is withheld and the reason given.

The ONS neighbourhood-level house price series ended in March 2023 and will not update. Median sale price and its year-on-year change now come from the ONS series for the wider neighbourhood group an area belongs to, current to September 2025, and each value says so.

The median and its change carry a confidence based on the number of sales behind them, so an inherited figure never reads more certain than its sample.

A year-on-year change built on a handful of neighbourhood sales is no longer shown.

Rail, metro and bus access now use authoritative national data with consistent coverage across England, Wales and Scotland. Transport scores no longer depend on how well an area happens to be mapped.

Amenity counts (schools, food and drink, healthcare, shops, parks and leisure) are precomputed for every area, so values are stable from one request to the next.

Median sale price and its year-on-year change now use an official rolling-year small-area price series instead of a self-computed figure that could swing on a handful of sales. Every value keeps its confidence rating.

Every preset now scores the same seven categories (safety and crime, deprivation, property, schools, amenities, transport, environment) from the full set of sources. The decision you are making only changes how they are weighted: property counts for more when investing, transport when choosing a business site, deprivation when moving.

Re-weight the same seven categories for any preset, so a score can follow your own priorities. Every response is still stamped with the engine version that produced it, so numbers stay reproducible for audits.

OneGoodArea now publishes a documented page for AI agents and a crawlable API description, so the AI assistants your team uses can read, query and cite us accurately. The scoring stays deterministic and version-stamped, so any answer they give reproduces.

The data and intelligence layer for UK area workflows, live. Four products on one shared store of signals and one fixed scoring engine.

Every public signal for an area across seven categories: crime, deprivation, property, schools, amenities, transport and environment. Each value carries its source, when it was measured, its confidence, and where it sits nationally and regionally.

A single 0 to 100 score for an area, built from the seven categories for one of four presets. The scoring never uses AI, so the same inputs on the same methodology version return the same score, and every result is stamped with the version that produced it.

Watch a list of areas and get told when something material changes, with a signed webhook. Small, noisy moves are held back, so you only hear about the ones that matter.

Ask in plain English or send a typed query. You get the answer and the exact query behind it, so every result can be checked and run again. Includes similar-area comparison, outlier detection and a straightforward forecast. The AI works out the query; it never decides the numbers.

Per-organisation signal bundles, saved scoring presets, peer groups, role-based access, white-labelling, and per-key IP allowlisting.