ATTOM、1億6,000万件の不動産記録を照会する3つのAIエージェントを発表

Why it matters

ATTOM is using proprietary property records to climb from data licensing into analysis and finished reports, while MCP keeps AI access governed, metered and tied to its APIs.

ATTOM CEO Rob Barber on August 18th introduced three specialized AI agents that let customers query property records, analyze housing markets and generate finished reports through natural-language prompts. The リリース extends ATTOM's data business into the work customers perform after retrieving a record, placing the vendor closer to underwriting, portfolio research and client delivery.

Barber has spent more than three decades in real estate information services. He joined ATTOM's predecessor, RealtyTrac, as CEO in 2015, led the creation of its consolidated data warehouse and oversaw the 2016 rebrand to ATTOM. Before that, he spent a decade running Environmental Data Resources, where ATTOM says he shifted a specialized property database toward workflow and market-intelligence products. The new agents follow the same playbook: use proprietary records as the base, then sell access to the decisions and documents built from them.

The technical work falls under Chief Product and Technology Officer Todd Teta, whose background puts him unusually close to the market ATTOM is trying to reshape. According to ATTOMのリーダーシップ略歴, Teta co-founded mortgage and real estate analytics provider VisionCore, sold it to CoreLogic and later led research and development groups that built CoreLogic's early mobile and mapping products. He studied computer engineering and computer science at the University of Southern California and joined ATTOM after leading product and technology at Meyers Research.

より狭い役割を持つ3つのエージェント

The Property & Place Insights Agent handles questions about individual properties, parcels, neighborhoods and geographic areas. ATTOM says a user can ask for property intelligence in natural language rather than selecting an API endpoint, assembling a query or navigating a conventional research interface.

The Data Analyst Agent works across aggregated records. ATTOM positions it for housing-market comparisons, trend analysis, portfolio evaluation and research queries. That distinction matters operationally. A property lookup typically starts with a known address or parcel, while a market query may require selecting geographies, time periods and measures before aggregating many records.

The Report Generation Agent pushes furthest into the application layer. ATTOM says it can assemble property characteristics, automated valuations, comparable sales, maps and neighborhood information into a customer-ready document. For a lender, brokerage, investor or insurance operation, that output could replace several steps between data retrieval and a report delivered to an employee or client.

ATTOM describes the agents as part of a self-service workbench built with the Model Context Protocol, or MCP, and the Agent2Agent protocol. Customers can also connect ATTOM Intelligence to applications that support those standards, including Claude and ChatGPT. ATTOM continues to support direct APIs, bulk licensing and cloud delivery, so the agents add another route into the same data estate rather than replacing ATTOM's established delivery products.

ATTOM says that estate covers 160 million U.S. properties, or 99% of the U.S. population. Its records span property characteristics, ownership, deeds, mortgages, foreclosures, valuations, environmental hazards, school areas, neighborhood information and geographic boundaries. The coverage figure and customer claims are supplied by ATTOM; ATTOM does not publish audited revenue, customer-count or usage figures for the new agents.

MCPサーバーがコントロールプレーンになる

ATTOM's agents sit on an MCP server that translates model tool calls into requests against ATTOM's existing systems. The server exposes structured tools for property profiles, ownership records, tax history, sales and mortgage history, automated valuation models, comparable sales, foreclosures, building permits, rental estimates and home-equity data.

That design gives developers a standardized interface while preserving the controls expected around licensed property data. ATTOM says access runs through authentication, plan entitlements, permissions, logging and usage monitoring. Its サポート文書 says the MCP server acts as a real-time broker and does not persist requests or responses. Traditional applications can continue calling ATTOM's APIs directly.

The implementation also carries familiar usage economics under the agent language. Each tool invocation counts as one request, regardless of how much information the tool returns. A workflow that separately calls a property profile, valuation model, comparable-sales tool and permit record therefore consumes four requests under ATTOM's accounting. Rate limits and monthly quotas are tied to customer plans, and ATTOM instructs developers building high-volume workflows to handle 429 Too Many Requests responses with retry logic.

ATTOM uses bearer-token 認証, which keeps the server recognizable to developers already operating API-based products. The abstraction removes some orchestration work, especially for low-code tools and internal assistants, while retaining the underlying mechanics of metered data access. That balance is central to ATTOM's commercial strategy: make the database easier for AI systems to consume without surrendering control over permissions, consumption or licensing.

The specialized agents represent a second layer above that interface. MCP gives a model a governed way to call property tools. ATTOM's agents package those tools around defined jobs, including answering a property question or producing a report. That can shorten deployment time for customers that want an outcome and have little interest in assembling their own prompts, tool sequences and output templates.

ソフトウェアスタック上での3段階の上昇

ATTOM has staged the rollout across 2026. On 1月27日, ATTOM introduced the MCP server alongside Databricks delivery. That release established an AI-focused connection to existing property endpoints and gave analytics teams another way to consume licensed datasets inside their cloud environments.

On 5月12日, ATTOM grouped its data, delivery systems and analytics under the ATTOM Intelligence name. The framework organized ATTOM's products into three layers: engineered data, AI-native delivery and analytics that produce decision-ready outputs. The August 18th agents give that framework visible applications rather than another taxonomy for the same catalog.

ATTOM also continues its work with Snowflake, where customers can access licensed datasets inside their own Snowflake environments. Cloud delivery and MCP address different integration points. Snowflake brings the records to a customer's data warehouse for internal analysis, while MCP lets an AI application retrieve specific records and analytical outputs at runtime.

The agents give ATTOM a direct role in the final workflow. That creates a useful commercial tension. ATTOM tells customers that it supplies data without competing against the products they build. A report-generation agent moves closer to functions that software vendors, brokerages and research platforms have historically implemented on top of ATTOM records. ATTOM can preserve that positioning if customers treat the agent as infrastructure embedded in their own products. A standalone ATTOM-generated report places the product further into customer-facing software.

Barberのプライベート・エクイティによるミッションがエージェント・サイクルと出会う

Lovell Minnick Partnersは2019年1月にRenovo CapitalとRosewood Private InvestmentsからATTOMを買収した。金銭面の条件は開示されていない。Lovell MinnickはATTOMの継続的なライセンス収入と追加市場への拡大機会を強調し、Barberは当時ATTOMがエンドツーエンドのデータプラットフォームを構築する計画だと述べた。

その後ATTOMは買収や追加の提供製品を通じて事業を拡大した。AIの展開は同じ基礎記録からの成長への別のルートを提供する。アクセスが容易になればクエリ量は増加し得る。パッケージ化された分析はより高付加価値の契約を支えることができる。完成したレポートはAPIやデータウェアハウスのテーブルを直接扱うことのない従業員にとってATTOMを関連性のあるものにする可能性がある。

リスクは実行時における信頼性にある。物件記録は、もっともらしい答えでは不十分なクレジット、保険、鑑定、投資、政府のワークフローに取り込まれる可能性がある。ATTOMの主張は、モデルの恣意性の余地を減らすために構造化されたツールと管理された記録を利用することに基づいている。レポートエージェントは基準をさらに引き上げる。というのも、磨かれた出力はアナリストの時間を節約するのと同様に、弱い情報源の選定、古い記録、あるいは不適切な評価を隠してしまう可能性があるからだ。

ATTOMのリリースはエージェントの独立した正確性率を示しておらず、1億6,000万件の物件という数値は回答の質ではなくデータベースのカバレッジを示している。製品の価値は、各応答がその基礎となる記録、日付、分析上の前提をどれだけ明確に示すかによって決まるだろう。プロパティインテリジェンスにおいては、出所（プロヴェナンス）は結果の一部である。

BarberはこれらのエージェントをATTOMの不動産データとAIへの長期的投資のさらなる一歩だと述べた。より厳密に読めば、会話型インターフェースが生のアクセスだけの価値を下げる中で、ATTOMは自社データベースの経済的地位を守ろうとしているということだ。もし顧客が物件のワークフローをプロンプトで始めることが増えるなら、ATTOMは記録を提供し、呼び出しを計測し、そこから生成される文書を出力したいと考えている。