<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Saurav Roy</title>
    <description>The latest articles on DEV Community by Saurav Roy (@easeweather).</description>
    <link>https://dev.to/easeweather</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4197954%2Fae56ee7c-4a9b-407f-a665-ba18ca9cbf32.png</url>
      <title>DEV Community: Saurav Roy</title>
      <link>https://dev.to/easeweather</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/easeweather"/>
    <language>en</language>
    <item>
      <title>How to Let ChatGPT or Claude Find a Local Service Provider with MCP</title>
      <dc:creator>Saurav Roy</dc:creator>
      <pubDate>Fri, 13 Nov 2026 08:40:05 +0000</pubDate>
      <link>https://dev.to/easeweather/how-to-let-chatgpt-or-claude-find-a-local-service-provider-with-mcp-4b36</link>
      <guid>https://dev.to/easeweather/how-to-let-chatgpt-or-claude-find-a-local-service-provider-with-mcp-4b36</guid>
      <description>&lt;p&gt;&lt;strong&gt;Subtitle:&lt;/strong&gt; A practical guide to connecting an AI assistant to a remote MCP server, searching by city, and sending an enquiry only after you approve the message.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tags:&lt;/strong&gt; MCP, ChatGPT, Claude, AI agents, local services&lt;/p&gt;

&lt;p&gt;People already ask ChatGPT and Claude to find a house cleaner, a personal trainer, or someone who can come on Saturday. Without a tool behind the request, the assistant is guessing from training data or from whatever it can browse. That is a weak way to reach a real local business.&lt;/p&gt;

&lt;p&gt;Model Context Protocol (MCP) is a standard way for an assistant to use tools. If you add a remote MCP server, the assistant can search a directory, read a public profile, and — only with your approval — send a message to a business. This guide walks through that setup using Servana's public MCP server for local services in the United States, and it is honest about what the server can and cannot do today.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the server actually does
&lt;/h2&gt;

&lt;p&gt;Servana is a booking platform for local service businesses. For AI assistants, it also publishes an MCP server for local service search. The official setup page is &lt;a href="https://servana.ai/for-agents" rel="noopener noreferrer"&gt;Servana for Agents&lt;/a&gt;. The server URL, which you paste into your assistant, is:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;https://servana.ai/api/mcp
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;No API key, no signup, and no approval wait. A GET request to that URL returns HTTP 405, which is expected: the endpoint speaks the MCP protocol over POST, not a normal web page.&lt;/p&gt;

&lt;p&gt;The server exposes five tools. The four read tools are documented as safe to auto-run. The fifth reaches a real business and asks for your approval first.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;search_services&lt;/code&gt; maps a free-text request to a service category. Example: "What Servana category covers someone to fold my laundry?"&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;search_vendors&lt;/code&gt; finds vendors by state, city, and optional service. Example: "Find cleaning services in Tampa, Florida on Servana."&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;get_vendor&lt;/code&gt; returns one vendor's public profile, services, and pricing. Example: "Show me the Servana profile for that first vendor."&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;estimate_service_request&lt;/code&gt; gives a heuristic price range from real vendor rates in that city. Example: "Roughly what would a 2 hour cleaning cost in Dallas on Servana?" Treat that figure as a rough range, not a quote.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;send_message_to_vendor&lt;/code&gt; sends an enquiry to a vendor's Servana inbox, with your approval. The business replies to you directly.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Servana does not process the payment or confirm the job. On &lt;a href="https://servana.ai/" rel="noopener noreferrer"&gt;Servana&lt;/a&gt;, the AI chat answers questions, then the customer sends a booking request, and the business accepts or declines. The MCP flow is the assistant-side version of that.&lt;/p&gt;

&lt;h2&gt;
  
  
  Coverage, honestly
&lt;/h2&gt;

&lt;p&gt;Servana is not a nationwide directory yet. As of November 12, 2026, the for-agents page lists these public listing counts in its deepest cities:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;City&lt;/th&gt;
&lt;th&gt;Vendors listed&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;New York, New York&lt;/td&gt;
&lt;td&gt;151&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Atlanta, Georgia&lt;/td&gt;
&lt;td&gt;108&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Jacksonville, Florida&lt;/td&gt;
&lt;td&gt;80&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Los Angeles, California&lt;/td&gt;
&lt;td&gt;74&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Dallas, Texas&lt;/td&gt;
&lt;td&gt;73&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Houston, Texas&lt;/td&gt;
&lt;td&gt;46&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Chicago, Illinois&lt;/td&gt;
&lt;td&gt;36&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Categories include cleaning and organizing, personal training, med spa, yoga, laundry, and more, depending on the city. Counts change as vendors join. Listings come from public business information and from owners, and some are still unverified, so confirm a profile with the business before you rely on it.&lt;/p&gt;

&lt;p&gt;If you search outside the cities above, the result may be thin or empty. That is the honest answer: there is no listed business there yet. Do not let the assistant invent one.&lt;/p&gt;

&lt;h2&gt;
  
  
  Connect it in Claude
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open Claude (claude.ai or the desktop app) and go to Settings, then Connectors.&lt;/li&gt;
&lt;li&gt;Choose "Add custom connector", name it Servana, and paste &lt;code&gt;https://servana.ai/api/mcp&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;In a new chat, switch Servana on from the tools menu, then ask in plain words.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Menu names can shift a little as the app updates.&lt;/p&gt;

&lt;h2&gt;
  
  
  Connect it in ChatGPT
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open ChatGPT settings, then Apps &amp;amp; Connectors. Custom connectors need a paid plan.&lt;/li&gt;
&lt;li&gt;Turn on developer mode under Advanced, choose Create, and paste the same server URL.&lt;/li&gt;
&lt;li&gt;In a new chat, add Servana from the tools menu, then ask in plain words.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Any other MCP client
&lt;/h2&gt;

&lt;p&gt;Paste this config into the client's MCP config file, or paste the URL into its connector settings:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"mcpServers"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="nl"&gt;"servana"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
      &lt;span class="nl"&gt;"url"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"https://servana.ai/api/mcp"&lt;/span&gt;
    &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Any MCP-capable client that can connect to a remote server URL should work.&lt;/p&gt;

&lt;h2&gt;
  
  
  A worked example
&lt;/h2&gt;

&lt;p&gt;Suppose you live in Jacksonville, Florida, and you want a two-hour house cleaning on Saturday. A useful sequence looks like this.&lt;/p&gt;

&lt;p&gt;First, ask the assistant to map the job to a category: "What Servana category covers a two-hour house cleaning?" That should call &lt;code&gt;search_services&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Next: "Find cleaning services in Jacksonville, Florida on Servana." That should call &lt;code&gt;search_vendors&lt;/code&gt;. Read the list. If it is short, that is useful information too.&lt;/p&gt;

&lt;p&gt;Then pick two or three listings and ask for each profile: "Show me the Servana profile for [name]." Look at services, area, and any published rates.&lt;/p&gt;

&lt;p&gt;If you want a rough sense of cost, ask: "Roughly what would a 2 hour cleaning cost in Jacksonville on Servana?" That is a heuristic from vendor rates in that city, not a binding quote.&lt;/p&gt;

&lt;p&gt;When you are ready to reach a business, draft the message yourself or edit what the assistant proposes. A clear enquiry includes the service, the neighborhood or address, the day and time window, how long you think the job will take, and how the business should reach you. Then say: "Send that cleaning vendor a message asking about Saturday availability." The assistant should show you the exact message and wait. Approve it only if you would be comfortable receiving it.&lt;/p&gt;

&lt;p&gt;The business replies to you directly. You still decide whether to book, how to pay, and whether the scope is right. The site names cash, Zelle, or Venmo as examples of paying the business directly; you and the provider agree on the method.&lt;/p&gt;

&lt;h2&gt;
  
  
  Habits that keep this from going wrong
&lt;/h2&gt;

&lt;p&gt;Ask in plain language, but include the city and state. "Find a cleaner near me" is weaker than "Find cleaning services in Dallas, Texas."&lt;/p&gt;

&lt;p&gt;Do not skip approval on &lt;code&gt;send_message_to_vendor&lt;/code&gt;. It reaches a real inbox.&lt;/p&gt;

&lt;p&gt;Ask about two or three businesses, not one, especially when you need a reply soon. Small providers often respond on their own schedule, and the tools will not invent availability they have not been given.&lt;/p&gt;

&lt;p&gt;If the assistant cites a business that the tools did not return, stop and have it search again.&lt;/p&gt;

&lt;p&gt;This is a search-and-enquiry workflow, not a checkout flow. The assistant does not hold a payment or lock in a time slot.&lt;/p&gt;

&lt;h2&gt;
  
  
  When this helps, and when it does not
&lt;/h2&gt;

&lt;p&gt;This setup is most useful when you already have an assistant open, you know the city, and you want structured results instead of a generic web roundup. It is less useful if you need a trade in a city with little or no coverage, or if you need to see the person and the workspace before you decide. In those cases, use the tools for a first pass, then confirm everything with the business.&lt;/p&gt;

&lt;p&gt;The setup steps, server URL, and tool list live on the for-agents page linked above. Start there if a menu name in Claude or ChatGPT has changed since this was written.&lt;/p&gt;

&lt;p&gt;Disclosure: I work with Servana.&lt;/p&gt;

</description>
      <category>mcp</category>
      <category>chatgpt</category>
      <category>claude</category>
    </item>
  </channel>
</rss>
