Repository navigation
💡 AI Answer-Engine Live Feed: Public Market Status API & MCP Server for ChatGPT, Claude and Gemini #548
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Summary
Publish a read-only, cache-friendly public feed of each market's verified, manager-published status: hours, which vendors are checked in today, and product exceptions, each with freshness timestamps. Offer it in three forms: schema.org Event/LocalBusiness JSON-LD, an llms.txt-style summary, and a hosted MCP server. AI assistants could then answer 'what's at the Riverside farmers market this Saturday?' from current, trusted data instead of stale web pages, and each answer would deep-link into the app.
Market Signal
According to secondary write-ups of BrightLocal's Local Consumer Review Survey (Feb 2026), 45% of US consumers used ChatGPT or other generative AI for local business recommendations, up from 6% a year earlier. AI is now the #3 local discovery source after Google and Facebook, and Google's share fell from 83% to 71%. Location-data vendors have shipped MCP servers so agents can read live business data: Yext, Uberall, PinMeTo, Local Falcon and Grid My Business (2026). Farmers markets are mostly static Facebook pages, so AI answers about them are usually stale. That is the failed-trip problem moved to a new channel.
User Signal
The PRD's core problem is customers acting on stale info, and its north-star metric is failed-trip reduction. The PRD also lists app-install friction as the main distribution risk for the mobile-only MVP. An answer-engine feed reaches people who haven't installed the app and sends them in via deep links. Related but different ideas: #479 (embed widget, aimed at human visitors to websites) and #539 (Google Maps hours sync, aimed at one platform). This feed is a single machine-readable source that any agent can read.
Technical Opportunity
Firebase Realtime already holds a denormalized, derived status projection that is not authoritative. A scheduled job in internal/scheduler can turn it into static JSON-LD and llms.txt files on a CDN at almost no cost. A small read-only MCP server would wrap the existing GraphQL public queries for listing markets, getting a market's day status, and finding vendors by product. It is a new adapter on the existing ports with no domain changes. Only manager-published data is exposed, so the PRD's trust and audit model stays intact. Expo Router's deep-linking means links in AI answers open the right screen.
Assessment
Adversarial Review
Strongest objection: The PRD explicitly deferred web and SEO for the mobile-only MVP. Assistants may not pick up the feed, and publishing data openly could help competitors scrape it.
Rebuttal: This is not a web app. It is a static data artifact plus an optional MCP endpoint, a few hundred lines with no UI, so the deferral doesn't apply. Market-day status is only useful when people see it, and markets already want it broadcast. Freshness timestamps make scraped copies go stale within hours, so Markets stays the source of truth. Usage is easy to measure: MCP calls and deep-link attribution show within one season whether assistants are using the feed. If they aren't, the static JSON-LD still improves Google results.
Suggested Next Step
Define a versioned public 'MarketDayStatus' JSON schema based on schema.org Event and Offer. Add a design note to architecture.md that names the Firebase projection as the only source and the published fields as the privacy boundary. Prototype a read-only MCP server against seed data and test it with Claude and ChatGPT connectors.
Generated by the weekly feature-ideation run (Mary, BMAD Analyst) on 2026-11-13. Sources: see Market Signal; figures from secondary write-ups are marked as such and should be checked against the primary source before you rely on them.
All reactions