us-property-parcel-real-estate-debt

Purpose

This is a service published by Capital-W-Holdings that provides AI agents with queries of U.S. real estate debt and property parcel data through MCP (Model Context Protocol). Its core goal is to enable agents to answer specific dated questions: who owns a given building, at what price it most recently sold, which loans and subsidies will mature within a given state and time window, and every conclusion comes with a data source and observation date.

Main Capabilities

The real estate domain contains two groups of data that barely overlap:

U.S. commercial and federal program real estate debt: loan maturity data covering 52 state codes, compliance maturity data covering 56 state codes, and subsidy maturity data covering 54 state codes. The official text explicitly lists 19,821 FHA and CMBS loans, 11,956 LIHTC expirations, and 4,721 HUD maturities. All loans carry maturity_basis = 'confirmed', meaning the dates come from filings submitted by loan servicers to the SEC or from HUD records and are resolved to specific buildings; not a single one is estimated or inferred from term.

Property records: 292,039 Massachusetts and New York parcels carry ownership and assessed value, and 95,562 recorded sales instruments cover the District of Columbia, Massachusetts, and New York.

The same connection also covers three other DFX graphs: family offices (2,559 offices, 4,238 foundations, 748 offices holding 13F positions, 394 offices with observed direct investments), independent sponsors (4,167 sponsors, 10,660 capital providers, 97,287 private companies, of which 17,064 carry transformation signals, 317,710 computed company-to-sponsor matches, 2,397 announced deals), and venture capital (76,936 funds, 113,590 people, 47,455 companies, 30,805 rounds). The four graphs use the same id scheme and the same response contract, and perform cross-graph identity linkage only through shared CRD, CIK, or EIN; entities with the same name are returned as candidates and are never merged.

How to Use

Endpoint:https://exchange-production-9123.up.railway.app/mcp Transport: Streamable HTTP Authentication: none Registry:io.github.Capital-W-Holdings/us-property-parcel-real-estate-debt

A plain GET to the endpoint returns the full tool list, coverage numbers, and a runnable example, with no handshake, session, or initialize required. Specific tools are then called via JSON-RPC. All 102 tools are free, unauthenticated, and read-only: no key, registration, or OAuth is required. If the client only supports stdio, you can use bridge/dfx_mcp_stdio.py as a bridge; that file depends only on the standard library, has no key, is stateless, and requires an absolute path when used.

The official documentation recommends reading the schema before calling, and recommends starting with what_can_dfx_answer ; that tool will explicitly say "no" and record the question, thereby allowing questions DFX cannot answer to influence the direction of subsequent builds. The server also rejects unknown parameters using an accompanying vocabulary.

Common tools:dfx_coverage returns the coverage grid when called;debt_maturity_schedule returns up to 200 loans for a given state and a forward window, one per line, sorted by maturity date, including principal, interest rate, lender, collateral property, and source fields;search_property_events returns events (date, title, address);get_property_record can reach loans that do not belong to a single state;search_entities、get_entity、search_people、search_relationships、relationship_path、search_events、verify and the economic tool find_capital_for_opportunity、find_opportunities_for_capital、explain_match、why_now、who_should_care；changes_since Sorted by cursor, by the time DFX learned the fact.

Use cases

• Query commercial mortgages, lenders, and corresponding buildings in a given state that mature within the next 548 days.
• Query where and when LIHTC compliance periods and HUD subsidy contracts expire.
• Query a parcel's last sale price, buyer, and registry book and page.
• Query a building's owner, manager, or lender.
• Perform entity retrieval, relationship path tracing, and match explanation across family offices, independent sponsors, and the venture capital graph.

Limitations

The official text explicitly lists several coverage unevennesses and limitations:

• Coverage is intentionally uneven and declared in advance: loan maturity coverage is only federal program loans (FHA-insured and agency-backed); county deed registries are not open to automation, and properties with traditional financing have no debt records here; a property not appearing in a maturity search does not mean the property has no debt.
• Massachusetts-exclusive: parcels and ownership. Recorded sales: all of Massachusetts, plus New York City deeds of $10 million and above. Boston-exclusive: permits and certificates of occupancy.
• LIHTC compliance periods are statutory, all falling on December 31, so daily bucketing produces a December cliff, which is an artifact of statute rather than a market event.
• No prediction in this graph has ever had an observed outcome; nothing carries calibrated probabilities, and every score is a ranking signal.
• A single street address may carry multiple records: among 6,114 such clusters, 2,764 have consistent unit counts and 3,350 have differing unit counts. DFX has not merged any of them,resolve() and will state which case applies.
• Property records and parcels are two almost non-overlapping populations, with only 661 clean one-to-one pairs out of roughly 100,000 each.
• Property records are not one row per building: 101,991 published property records cover 94,859 distinct normalized addresses, and computing totals across them overestimates by about 8%; each record carries address_group_size to facilitate deduplication.
• Sales data comes from two sources with different granularity: Massachusetts is an assessment roster (statewide, one sale per parcel, buyer named, seller never named, so repeat-sale pairs and holding periods cannot be derived), and New York is a registry extract (five boroughs, both parties named, every instrument dated, so repeat sales and holding periods can be derived, but only for deeds of $10 million and above). Neither is nationwide sales data.
• 145 loans have no single state attribution and cannot be reached through state filtering; they must be obtained through get_property_record .
• Contact points are not available via MCP on any graph.
• Some venture capital fields in the official text display as None (such as fund count, number of institutions with recent fundraising, and number of institutions with filled industries); the specific meaning is not yet provided.

Page data is labeled as measured on 2026-09-30, and it is recommended to call dfx_coverage to get the same grid at reading time.