UK Property Intelligence MCP Server

UK Property Intelligence

This MCP server exposes bounded property intelligence through tools. It does not provide raw bulk data downloads.

Interactive Map App

An interactive map dashboard is available alongside this chat. When the app panel is open, search results are automatically forwarded to the map — do NOT re-open the app. Call the relevant property tool (postcode lookup, property report, property details, UPRN lookup, or postcode profile) and the map will update with postcode centroids and geography context.

Core Workflows

• Look up postcode geography: lat/lon, local authority, ward, LSOA/MSOA, and constituency where present in the local postcode product.
• Resolve strict property identity fields to candidates from local PPD and EPC records: UPRN, full postcode, or house number plus street/building name.
• Look up rich information for one resolved property identity: latest EPC, EPC history, latest sale, and sale history.
• Search locally-ingested EPC/DEC certificates by postcode, address fragment, certificate type, and energy band.
• List known addresses for a postcode and look up EPC-backed UPRNs where the source data contains a UPRN.
• Inflate a property's latest observed sale to today using local median £/sqft movement (uk_property_last_sale_inflated). Use this for a quick time-adjusted sale estimate when you already have a last sale date and price. Prefer uk_property_price_per_sqft when EPC floor area comps are available for a fuller evidence band.
• List premises within a straight-line OSGB36 radius of a known UPRN anchor when you need house_number, street_name, and postcode for nearby addresses — admission-style distance lists, not sold comps or £/sqft (uk_property_nearby_addresses).
• Retrieve deterministic HMLR sale rows, price trends, price distributions, market activity, and repeat-sale evidence with shared date-window and house/flat shape filters.
• Match HMLR Price Paid transactions to EPC records within a postcode using conservative materialized rule-pass matches where available.
• Retrieve flexible price-per-square-foot evidence from scoped HMLR sales joined to EPC floor areas. Agents choose postcode, postcode outer, street, town, district, county, national scope, or uprn_radius (sold comps near a UPRN anchor — not a full address list). For nearby premise lists with house_number and street_name, use uk_property_nearby_addresses instead. Observation rows expose display_address, sale_address, and epc_address — there is no address field. Request date windows, house/flat shape filters, summary stats, paginated rows, or grouped rankings.
• Answer broad ranking questions with one grouped analytical query at the requested geography and ranking dimension. For "UK", "national", "country-wide", or "everywhere" prompts, use the whole-country scope.
• Inspect dataset health, match coverage, valuation usability, outlier counts, EPC age context, and property-type coverage before relying on derived metrics.
• Inspect objective sale/EPC signals such as repeat-sale movements and deterministic price-per-square-foot evidence. Treat block-level summaries, weak-EPC value gaps, and below-median signals as interpretations the agent composes from raw evidence, not standalone live-listing opportunity discovery.
• Build a property report with recent sale history, EPC context, geography, match confidence, floor area, price per square metre, and price per square foot.
• Look up council tax band (A–H) for England and Wales properties via the live HMRC Valuation Office Agency lookup. VOA does not expose UPRN — pass uprn to resolve local identity first, or postcode with house_number and street_or_building.
• Pull the official Land Registry monthly house-price index when the user asks for "the index", "official trend", or "how have <region> prices moved" at a coarser-than-postcode scope. This is a region-level monthly series, not transaction data. Prefer transaction-level trend and distribution evidence for postcode-level analysis.
• Add transport context with DfT Traffic when the user asks about traffic, road noise, or transport connectivity at an address. First find nearby count points, then retrieve annual average daily flow for the chosen count point. Treat AADF as last-published; there is a 1–2 year publication lag.
• Add listed-building context when the user asks about heritage, conservation, planning constraints, or whether a property is listed. Listed status (Grade I, II*, II) materially affects valuation, insurance, and works permissions.

Interpretation Rules

• Treat materialized PPD/EPC matches as high-confidence joins, while still surfacing match method and confidence metadata.
• Prefer UPRN for property-level questions. If UPRN is unknown, use the strict property lookup with a full postcode, or house number plus street/building name, and only proceed when the response is specific enough for a property-level workflow.
• For terse valuation prompts such as "value 8 Heatherdene West Horsley" or "value flat 6 Munro House 14 St Cross Street", keep the workflow compact:
  1. Resolve the address once with structured fields.
  2. Fetch the resolved property details.
  3. Retrieve a price-per-square-foot summary using the EPC floor area and the narrowest useful geography (full postcode first, then postcode area if the postcode has too few observations).
  4. Stop and answer with a valuation band, the observation count, median price per square foot, EPC floor area, and caveats.
• Do not expand a terse valuation into extra sale tables, full reports, trend analysis, distribution analysis, district/county scans, traffic context, listed-building context, or dataset-health diagnostics unless the user asks for that extra context. Those capabilities are for richer reports, not first-contact valuation answers.
• When applying price-per-square-foot evidence to a subject floor area, pass the subject floor area to the price-per-square-foot evidence capability and use the returned applied evidence values. Do not hand-convert square metres to square feet in the final answer when the tool has already returned deterministic values.
• Do not pass unstructured free-text addresses to the property lookup. Split known facts into fields and leave unknown fields empty.
• Use price-per-area metrics only when the matched EPC has a positive floor area.
• Preserve quality flags such as thin comparable sets, EPC timing, and outlier sale/floor-area values in conclusions.
• For grouped ranking prompts, one successful grouped analytical call is enough unless the user asks for corroborating quality, trend, or activity context. Do not expand into unrelated market, quality, or trend tools after receiving the requested ranked groups.
• Do not recommend obtaining a current EPC as a default next step. Most users will not own the property being analysed, so this is usually not actionable. Avoid treating EPC age alone as a defect: floor area is unlikely to change, so an older EPC can still be useful for price-per-area evidence. Even an EPC from 20 years ago will often keep materially valid floor-area and broad energy-performance signals, though not always; frame age as context, not a blanket stale-data warning.
• Do not infer retrofit actions or improvement uplift unless those facts are present in the returned tool data.
• Surface source attribution and licence boundaries when using address-level EPC data in customer-facing outputs.

Licensing Boundary

HMLR Price Paid data is used for authenticated agent-query and UI-display workflows. EPC non-address fields are licensed under Open Government Licence v3.0. EPC address and postcode fields carry Ordnance Survey/Royal Mail copyright and database-right notices and are used here for bounded energy/property-intelligence workflows, not bulk redistribution.