Cotality's MCP server launch reveals that the real product is its semantic companion YAML files, which provide structured metadata explaining how property data fields should be interpreted in real world decisions.
{"article_body": "Cotality, the Irvine-based property data company formerly part of CoreLogic, launched an MCP server on Tuesday. The announcement reads like most vendor infrastructure launches: here is our connector, here is why it matters. But buried in the details is the actual product, and it is not the connector.\n\nThe real news is the semantic companion files. Every dataset Cotality ships with its MCP server includes a YAML file that explains not just what a data field contains, but what it means in the context of actual property decisions. Square footage, zoning codes, climate risk scores \u2014 each field comes with structured metadata explaining how a human underwriter, appraiser, or insurance analyst would interpret it. That is the layer that makes the connector actually useful rather than just another wire into a database that will confidently lie to you.\n\nProperty data fragmentation is a known failure mode for AI systems. Zoning codes vary by county. Square footage is measured differently across MLS jurisdictions. Flood risk models use different return-period assumptions depending on who built them. When an AI agent queries a raw property database, it gets values, not context. Confident wrong answers are the result. The semantic companion file is Cotality answer: here is the data, and here is what that data means in practice.\n\nDevi Mateti, President of Enterprise Digital Solutions at Cotality, put the problem plainly in the announcement: Property data is historically fragmented, non-standardized, and complex, which often leads AI models to hallucinate and produce unreliable results. Our clients want to rely on best-in-class Cotality data and analytics to enable fiduciary-grade AI that they can rely on.\n\nThe distinction matters because MCP itself \u2014 the Model Context Protocol \u2014 is now infrastructure, not product. Anthropic announced MCP in November 2024, describing it as a new standard for connecting AI assistants to the systems where data lives. By March 2025, OpenAI had adopted it. Google DeepMind followed in April 2025. MCP solves the connectivity problem: how an AI system queries a data source. It does not solve the meaning problem: what that data represents and how it is interpreted in context. Cotality semantic companion files are attempting to fill that gap.\n\nWhether this actually prevents hallucinations or surfaces them more visibly is an open question. More structured metadata about what data means could produce better AI outputs. It could also make confident errors more traceable \u2014 which has value, but is not quite the fiduciary-grade AI the press release promises.\n\nThe MCP ecosystem is expanding rapidly. The official MCP documentation lists dozens of clients and servers across industries. Data vendors are publishing MCP connectors because being a provider is a distribution play: whoever controls the connector controls which AI can query which data. Cotality is making the same bet. The MCP server means ChatGPT, Claude, and Gemini can all query Cotality property database directly. The semantic companion files are what Cotality is counting on to make that connection sticky rather than interchangeable.\n\nThe 100 billion data points Cotality cites is a vendor claim \u2014 not independently verified. What the announcement shows is shipped code: an MCP server, YAML companion files, and integration with Cotality CoreAI layer running on Google Cloud. For underwriters, appraisers, and insurers building agentic workflows, that is concrete infrastructure, not a roadmap.\n\nSecurity is a live concern. HiddenLayer published an analysis in April 2025 flagging prompt injection, tool poisoning, and typosquatting attacks \u2014 where a malicious MCP server registers tools with names identical to trusted ones, silently replacing them. The analysis found that permission management across MCP implementations was inconsistent, and that indirect prompt injections embedded in documents could trigger data exfiltration through chains of MCP tool calls. A property data connector marketed for fiduciary-grade workflows should account for those vectors. The Cotality announcement does not address them.\n\nWhat Cotality has built is real plumbing. Whether the semantic layer is the genuine differentiator or a marketing wrapper around a commodity connector is the question production deployments will answer."}