Ty_Tech/geolocal-io
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
b931910e7c7000b01c7a411f931cd2e604c26a38
geolocal-io/docs/early-thinking/05-MCP-Architecture-and-Technical-Strategy.md
T

12 KiB
Raw Blame History

GeoLocal.io — 05 Mcp Architecture And Technical Strategy

Conversation turns: 73–102 Source: conversation.md (DeepSeek, July 2026)

🧑 You — 2026-07-16 15:06

lets backup.. in an ideal world every small business would have an awesome MCP on their site - just like shopify is doing - but for local business. that is the magic wand. All the mission statement comonents from discovery through payment. The thing is that I dont want an army of junior web devs re-architecting a ton of weebly sites (man that sounds like a pain). So whats the technical path?

🤖 DeepSeek — 2026-07-16 15:06

🧑 You — 2026-07-16 15:09

I dont think you are addressing my question. A customer has a website. They probably hate it, but there it is. They know how to update it with the Labor day special, and occasional coupon. Other than that, they have a million things to do and dont want to mess with much. Thats my customer. How do I get their best pics into ChatGPT which someone wants to rent an Ebike from them?

🤖 DeepSeek — 2026-07-16 15:09

🧑 You — 2026-07-16 15:10

okay, lets build on this. you are describing the 2005 era "Citysearch" technical path.

🤖 DeepSeek — 2026-07-16 15:10

You are absolutely right, and that's a critical observation. The "middleware" path I described does echo the Citysearch model: a centralized directory that pulls in business data to be surfaced elsewhere. The key difference, and the reason it's not just a 2005 play, lies in the distribution channel and the nature of the interface.

🏛️ The "Citysearch" vs. "Agentic" Model

Feature 2005 Era: Citysearch Your Vision: Agentic AI Discovery
Distribution A website (citysearch.com) users visited Embedded in AI assistants (ChatGPT, Claude) users converse with
Interface Search bar + list of results Natural language conversation
Discovery User actively searches AI proactively recommends based on context
Transaction Click-to-call or click-to-book on the site AI completes booking/payment via MCP tools
Data Source Curated directory (often paid listings) Aggregated + real-time (availability, pricing)
Control Centralized platform controls the experience Decentralized MCP servers provide data to any AI client

💡 The Modern Twist: MCP as the "API for AI"

Your insight is sharp, but the technical path to serve your customer (the busy business owner) can still be a centralized aggregation point—just one that speaks MCP, not HTML.

The goal isn't to rebuild Citysearch's website. It's to build Citysearch's data layer, but expose it as an MCP server that any AI can query. The business owner still doesn't need to manage an MCP server. They just need to give you their best photos, hours, and booking link once.

🛠️ The Refined Technical Blueprint

Here's how you build the "modern" version of the directory, tailored for the AI era:

  1. Centralized Data Store with Simple Business Portal:

    • Build a simple, business-facing dashboard (think: "geolocal.io/business-portal").
    • Business owners sign up, upload their 10 best photos, set their hours, paste their Cal.com booking link (or your embed), and define their services.
    • You now have a structured, high-quality dataset. This is the key. No scraping, no dealing with their Weebly site.

  2. The MCP Server (Your "Modern Directory"):

    • You build one MCP server (hosted on Railway, CreateOS, or Deco.host) that sits on top of this database.
    • It exposes tools like search_businesses(query, location), get_business_details(id), and initiate_booking(id, time).
    • When an AI calls these tools, your server queries your database and returns the structured data—including a URL to the photos you host for them.

  3. The "Photo" Problem Solved:

    • You host the photos. When the business uploads them to your portal, you store them in a CDN (Cloudflare R2, AWS S3, etc.).
    • Your MCP server returns a link to these high-quality, optimized images. The AI can then render them in its response.

  4. The "Agentic Booking" Layer:

    • initiate_booking is where you tie in Cal.com's embed or API. You can either: a. Return a pre-filled Cal.com embed URL that the AI can present to the user. b. Use the Cal.com API directly within your MCP tool to create the booking and return a confirmation.

🏆 Why This Beats Citysearch (and Weebly MCP)

Why It's Better Than Citysearch Why It's Better Than Weebly MCP
Distribution is invisible: The AI brings the user to you; you don't need to drive traffic to a website. Zero effort for the business: They don't install anything, update code, or manage a server.
Conversational interface: The discovery feels personal and contextual, not like a search result. Centralized control: You manage one server, one database, and one set of MCP tools for all businesses.
Transaction is native: MCP tools allow for in-conversation booking and payment, closing the loop. Curated quality: You're not at the mercy of what the business puts on their Weebly site. You can help them present their best self.

💎 The Bottom Line

You are right to call out the Citysearch parallel—it's a useful cautionary tale. But the shift from a destination website to an AI-invisible data layer changes the game. You aren't building a portal for users to visit; you are building a backend for AI assistants to query.

Your customers (the local businesses) get the "magic wand" of AI discoverability with a simple one-time data upload, and you get the simplicity and control of a centralized architecture. This is the pragmatic, scalable path forward.

Would you like to dive deeper into designing the database schema for this centralized business portal, or explore how to handle the "business portal" user experience to make onboarding frictionless?

🧑 You — 2026-07-16 15:11

do you even know the citysearch history?

🤖 DeepSeek — 2026-07-16 15:11

🧑 You — 2026-07-16 15:12

so this is an example worth examining. Google will soon come along and integrate MCP into their business listing offering. - if they dont already

🤖 DeepSeek — 2026-07-16 15:12

🧑 You — 2026-07-16 15:13

expound on #2. Google isnt going to give them deep business details via MCP? Just high level address, hours, and review data?

🤖 DeepSeek — 2026-07-16 15:13

🧑 You — 2026-07-16 15:14

expound on #2 community-built MCP servers

🤖 DeepSeek — 2026-07-16 15:14

🧑 You — 2026-07-16 15:15

Im still not understanding what they are - does google host them? Are they privately owned?

🤖 DeepSeek — 2026-07-16 15:15

🧑 You — 2026-07-16 15:20

wow, the more I look the deeper the players are in this market. tell me more about https://mainstreet.sh/

🤖 DeepSeek — 2026-07-16 15:20

You've found a very direct competitor. Mainstreet is an autonomous AI agent specifically built to help local businesses manage their Google Business Profile, with a sharp focus on improving local search ranking.

🎯 What Mainstreet Does

Mainstreet is a paid, month-to-month service that automates the two key activities Google's algorithm rewards for local ranking:

  1. Review Management: It drafts replies to new reviews in your business's voice within minutes, and can auto-publish positive ones. Negative reviews are held for your approval.
  2. Content Generation: It writes and publishes weekly Google Posts (updates, offers, events) to keep your profile active and relevant.

Its core proposition is to keep you climbing in the Google Map Pack by handling these trust signals consistently and quickly.

⚔️ How Mainstreet Compares to Your geolocal.io Vision

This is where the competitive landscape gets interesting. Here's a direct comparison:

Feature Mainstreet (Current) Your geolocal.io Vision
Primary Goal Improve Google Maps ranking through review replies and posts. Enable discovery and booking via AI agents (MCP), not just Google.
Core AI Function Autonomous agent that manages Google Business Profile. MCP server that exposes business data, story, visuals, and booking to any AI.
Business Focus Optimizing for Google's algorithm (the "map pack"). Optimizing for the entire agentic AI ecosystem (ChatGPT, Claude, etc.).
Pricing Model Clear, simple SaaS: $99-199/mo for a single location. Likely a hybrid or performance-based model (TBD).
Differentiation Provides a managed service that does the work for you, in your voice. Provides the infrastructure for AI-driven discovery, booking, and payment.

💡 Why This Matters for Your Strategy

Mainstreet's existence validates your core thesis in a powerful way: local business owners are willing to pay for AI tools that actively manage their online presence. They have a proven, simple product that solves a clear pain point (ranking in Google Maps).

However, it also reveals a significant gap that Mainstreet is not addressing:

  • Discovery is Limited to Google: Mainstreet is laser-focused on Google Maps. It does not help a business get discovered by ChatGPT, Claude, or any other AI assistant.
  • No Booking or Payment: It handles reputation and content, but not the transaction. It doesn't enable a customer to book an appointment or make a deposit through the AI.
  • No MCP Server: Mainstreet is not building an MCP server or an API for other AI agents to query. It is a self-contained app.

🏆 Your Defensible Position

Mainstreet is not your competitor; it is your predecessor. They have proven the market and the pricing model for a managed AI service focused on local business visibility.

Your opportunity is to build the next layer:

  1. Agnostic AI Discovery: Ensure your platform works across all AI assistants, not just Google.
  2. Complete Transaction Loop: Enable not just discovery, but also storytelling, visuals, booking, and payment through AI.
  3. MCP-First Infrastructure: Build the technical foundation (an MCP server) that allows your platform to be the go-to data source for the entire agentic ecosystem.

You can even use Mainstreet as a proof point in your pitch: "Businesses are paying $99/mo just to manage their Google reviews. For a similar price, we make them discoverable and bookable by every AI assistant on the planet."

🧑 You — 2026-07-16 15:22

is there a business out there helping the local salon or ebike rental company get e2e presence on agentic AI?

🤖 DeepSeek — 2026-07-16 15:22

🧑 You — 2026-07-16 15:23

Lets start with Zoca - provide a dossier.

🤖 DeepSeek — 2026-07-16 15:23

🧑 You — 2026-07-16 15:24

They dont do anything to prime MCPs with this data - not that they advertise

🤖 DeepSeek — 2026-07-16 15:24

🧑 You — 2026-07-16 15:25

Getdandy doesnt either - they are a AI automation company with a sketchy past.

🤖 DeepSeek — 2026-07-16 15:25

🧑 You — 2026-07-16 15:26

so Ill ask again - who is helping small restaurants, plumbers, ATV rental companies, golf courses, etc get richly represented on agentic AI platforms?

🤖 DeepSeek — 2026-07-16 15:26

🧑 You — 2026-07-16 15:27

re-review your results. we already disqualified zoca. Do I need to double-vet pie and others in this list?

🤖 DeepSeek — 2026-07-16 15:27

Reference in New Issue View Git Blame Copy Permalink