Markets render faster than models can read them.

The connector is a remote MCP server at mcp.maatsii.com/mcp: streamable HTTP, no authentication, 10 tools. It works in any MCP client, including Claude Code, Claude Desktop, ChatGPT, Cursor and Hermes.

COPY$ claude mcp add --transport http maatsii https://mcp.maatsii.com/mcp
Added HTTP MCP server maatsii

COPY{
  "mcpServers": {
    "maatsii": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.maatsii.com/mcp"]
    }
  }
}

COPY{
  "mcpServers": {
    "maatsii": { "url": "https://mcp.maatsii.com/mcp" }
  }
}

COPY$ curl -s https://mcp.maatsii.com/healthz
{"ok":true,"version":"1.5.1","auth":"none","charts":200,
 "setups":{"total":64,"fan":45,"tick":19}}

Then ask: Read NQ on the 15-minute and tell me where the fan sits.

Every frame is a public JPEG, c1 through c200, redrawn every 3 seconds with the fan, the rails and the price box on it. Four rules, then it works from anything that can fetch a URL.

$ curl -o nq.jpg "https://api.maatsii.com/c1.jpg?raw=1&cb=$(date +%s)"
200 · image/jpeg · cache-control: no-store · x-frame-timestamp on every frame

Build a URL

Machine-readable surfaces

Your agent can integrate Maatsii without you reading this page.

Read llms.txt →

64 active linear-regression setups run against the live catalog: 45 fan, 19 tick. Every one is drawn on the frame your model reads.

We publish the counts. We do not publish the rulebook.