Atlas Insight

Connect your AI

Put your own AI assistant on 14.5 million car-owner posts

Add one address to the assistant you already use: Claude, ChatGPT, Cursor or any assistant that accepts connectors (MCP). It can then count and quote 14.5 million posts from Vietnamese car forums; 13.5 million are analysed by topic.

The address to add

https://atlas-mcp.vad37.com/mcp

It is free, needs no login, and answers about 30 questions a minute. Two minutes on Claude; ChatGPT needs developer mode on some plans. The address shows nothing in a browser.

Claude

  1. On claude.ai or Claude Desktop, open Settings, Connectors.
  2. Choose Add custom connector, name it “atlas” and paste the address.
  3. In a new chat, start your question with “Using the atlas”.

Claude Code, in a terminal:

claude mcp add --transport http atlas https://atlas-mcp.vad37.com/mcp

ChatGPT

  1. Open Settings, Connectors. Some plans call this developer mode; turn it on if asked.
  2. Add a custom connector and paste the address.
  3. In a new chat, turn the connector on and ask.

Cursor

  1. Open Settings, MCP, and choose Add new MCP server.
  2. Paste this into the file it opens, then ask in the chat panel:
{"mcpServers":{"atlas":{"url":"https://atlas-mcp.vad37.com/mcp"}}}

Any MCP tool

  1. Find where your tool adds an MCP server or connector.
  2. Paste the address. If it asks for a type, choose “streamable HTTP”.
  3. Start your question with “Using the atlas”.

What is MCP?

MCP is a standard way for an AI assistant to read a collection like this one. Think of it as a library card for your assistant.

You add the address once. After that, the assistant looks things up when your question needs it and shows where each answer came from. You do not need to learn a chatbot of ours.

First questions to ask

  • Using the atlas, what do car owners complain about most after an insurance claim? Count the posts by year and quote five with links.
  • Using the atlas, which insurers do car owners name most, and which draw the most complaints per post? Show a small table.
  • Using the atlas, what do car owners believe their mandatory third-party insurance covers? Quote five posts where they are unsure, with links.
  • Using the atlas, who do car owners ask before they buy car insurance: friends, repair shops, dealers, agents or apps? Count each.
  • Using the atlas, what do car owners say to do in the first hour after another car hits them? Quote five posts.

More questions, with what they found

In a decision meeting

  1. Bring the brief's claim in one sentence, for example “car owners want their insurance papers in an app”.
  2. Ask your AI assistant to count posts by car owners for and against it, by year, without sellers and ads.
    Using the atlas, test this belief: car owners want their insurance papers in an app. Count posts by car owners that support it and posts that contradict it, by year. Leave out sellers and ads.
  3. Ask for five posts to read on each side, and how sure the AI assistant is.
  4. Decide: build it, drop it, or test it with real users first. The posts stay linked, so anyone can check later.

What it can and cannot tell you

It can

  • Say how many car owners raise something, and since when.
  • Give real posts, with links, for any answer.
  • Show whether a belief holds up in what car owners write.
  • Compare insurers by how often car owners complain.
  • Read our research notes and the sources behind them.

It cannot

  • Say whether what a car owner wrote is true; it counts what was written.
  • See social apps or forums outside the three it holds.
  • Quote prices or terms from insurers' own systems.
  • See posts written after October 2026, when we last collected.

Posts are in Vietnamese; your assistant translates as it answers. For engineers: index.json and llms.txt.