Mireye's API and MCP server return cited parcel, hazard, utility and site-selection data across the US.
By RuntimeWire Staff · Published
Primary source: Mireye
Why it matters #
Mireye is betting the valuable layer in physical-world AI will be the provenance and normalization that make messy public and licensed data usable by agents.
Ansh Chokshi and Shashwat Kapoor are building Mireye around a straightforward premise: AI agents cannot make reliable decisions about physical locations when the necessary facts are scattered across government databases, utility portals, maps and county records.
Mireye entered Y Combinator's Summer 2026 batch with a team of two and a product aimed at developers building software for land diligence, infrastructure planning, insurance, energy and site selection. Mireye packages fragmented public and licensed location data behind an API and MCP server, with returned fields carrying source, fetch-time and confidence metadata. Model Context Protocol, introduced by Anthropic as an open standard, connects AI assistants to external systems and data.
Chokshi arrived at the problem after leading AI and machine-learning engineering at sales software company Seismic. YC's profile says he studied computer science, mathematics and economics at the University of Toronto and previously co-founded Poker Pit, which reached $250,000 in annual recurring revenue within four months before an acquisition. The acquirer and transaction terms are not identified in Mireye's public materials, so the result remains Chokshi's account through YC.
Kapoor's public profile identifies Mireye as his current company and lists the University of Maryland, College Park. His role is central to the product's premise: Mireye must turn incompatible location records into consistent fields that an agent can query repeatedly.
Building the location layer under the agent
Mireye's API documentation covers endpoints for natural-language questions, geocoding, parcel lookup, structured field retrieval, proximity analysis and requests for new fields. In Mireye's API playground, a developer can test calls and copy the resulting integration in cURL, TypeScript, Python or MCP configuration.
The distinction Mireye is trying to establish is provenance. Its current homepage says responses include a source URL, fetch timestamp and confidence rating for each returned value. A site-selection agent could ask whether a property is suitable for a data center and receive fields covering flood exposure, slope, land cover and nearby power infrastructure, rather than relying on a language model's memory or an uncited web summary.
Mireye says its catalog draws from 85 authoritative sources, primarily federal datasets, county records and open datasets. The company names US agencies including the US Geological Survey, FEMA, NOAA, USDA, the Bureau of Land Management, the Environmental Protection Agency and the Energy Information Administration, along with sources such as Overture and Sentinel-2. Mireye is selling normalization, provenance and an agent interface over those records. It does not claim to own every underlying record.
On Mireye's YC profile, the company describes today's agents as "book smart" and its intended product as the layer that makes them "street smart." The phrase identifies a practical limitation. A model can reason about a parcel only after another system resolves the address, identifies the jurisdiction, finds the relevant records and explains how current and reliable each value is.
How missing fields enter the catalog
Mireye's August YC launch page described an on-demand indexing API powered by a long-running agent. When a requested field was missing, Mireye said the agent would source and index it for future users.
The current pricing FAQ says Mireye's team researches, builds and validates a requested field, then notifies the requester when it goes live. Mireye's API documentation says the field-request endpoint returns an ID that customers can use to poll the request's status. The two descriptions leave unclear how much of the current indexing process is performed by the long-running agent described in the YC launch materials.
Mireye's published catalog counts also differ. The YC launch materials claim more than 300 fields and 20% week-over-week catalog growth, while the current pricing page lists 285 cited, typed facts about a location. Mireye does not explain the difference between those figures. The growth rate and Mireye's claim that early customers sourced off-market data-center land 100 times faster have not been independently substantiated, and the customers are unnamed.
Cheap access is part of the wedge
Mireye's pricing page offers 5,000 credits each month at no charge. Its Build plan costs $19 a month for 25,000 credits, while Growth costs $99 for 120,000 credits. Enterprise pricing is negotiated.
The same pricing documentation lists geocoding at one credit and each field at one credit per location. Mireye prices a 90-field data-center siting screen at 90 credits, or about 9 cents, and a parcel lookup at 300 credits. Plans stop processing requests when the allowance is exhausted rather than automatically generating an overage bill, an important control when autonomous software can produce far more API traffic than its operator intended.
YC is Mireye's only publicly identified backer. Participation in the batch carries YC's published $500,000 standard deal: $125,000 for 7% and $375,000 through an uncapped SAFE with most-favored-nation terms.
Mireye is entering an occupied map
Location data already supports a mature software market, and existing providers are moving their products into agent interfaces. Regrid's MCP server, for example, exposes more than 160 million US parcel records covering boundaries, ownership, assessed values, land use and building footprints.
EagleView moved further up the application stack in April with Horizon, an agentic geospatial product built on its aerial imagery and property intelligence. RealEstateAPI offers property data through conversational and MCP interfaces, while LandTech documents APIs covering parcels, ownership, planning, constraints and power data.
Chokshi and Kapoor are betting that developers will prefer one agent-ready layer spanning several categories of location information, complete with citations and consistent schemas, over a stack assembled from separate specialist APIs. Mireye's field-request system gives that bet a compounding mechanism: each validated customer request can become a reusable part of the catalog.
Upstream datasets will change as customers use Mireye's answers for property, underwriting and infrastructure decisions. Its citations, timestamps and confidence fields give developers the raw material to audit an answer, while the field-request process gives Mireye a way to expand the catalog in response to demand.
Mireye starts with the right unit of work: a decision about a specific place. Chokshi and Kapoor have built the connective tissue between that question and the records needed to answer it. Their next test is whether developers want a general physical-world layer, or whether buyers continue purchasing location intelligence one specialized dataset at a time.