Use our test results in Claude, ChatGPT and Cursor

It's free, read-only and needs no account or API key. You add one address to your assistant:

https://modeltested.com/mcp

It covers all 67 models we've tested and is updated every time we publish new results (last update: Oct 18, 2026).

Claude (web, desktop and mobile)

1. Open Settings → Connectors and choose Add custom connector.
2. Name it Model Tested and paste https://modeltested.com/mcp as the URL. Leave the advanced settings empty.
3. In a chat, turn on Model Tested from the tools menu, then ask away.

On Team and Enterprise plans, an admin may need to add the connector for the organization first.

Claude Code

claude mcp add --transport http modeltested https://modeltested.com/mcp

ChatGPT, Cursor, VS Code and others

Any app that supports remote MCP servers over HTTP works with the same address.

• ChatGPT: turn on developer mode under Settings → Apps & Connectors → Advanced, then create a connector with the URL above and no authentication.
• Cursor: add this to ~/.cursor/mcp.json:

  {
    "mcpServers": {
      "modeltested": { "url": "https://modeltested.com/mcp" }
    }
  }

• VS Code: add this to .vscode/mcp.json:

  {
    "servers": {
      "modeltested": {
        "type": "http",
        "url": "https://modeltested.com/mcp"
      }
    }
  }

• Apps that only run local servers: use npx mcp-remote https://modeltested.com/mcp as the command.

What to ask

• "What's the best model I can run on a 16 GB graphics card?"
• "Which local model is best at coding on a Mac with 32 GB?"
• "Compare Qwen3.5 9B and Ornith 1.5 9B."
• "How much memory does gpt-oss-20b need at Q8?"
• "Which decision model is the most accurate, and what does it cost?"

Tools

Tool	What it returns
find_models	The best models for a task (overall, coding, documents, decisions, hard tasks), optionally only those that fit a graphics card or Mac of a given size, with the download to use and the memory it needs.
get_model	Everything we have on one model: scores, rank, memory at each quantization, the smallest GPU and Mac it fits, decision accuracy and our Q4 run on a Mac.
compare_models	Two to six models side by side.
decision_models	Decision model rankings: accuracy, hard tasks, how honest their confidence is, and cost per 1,000 calls.
search_articles	Our articles and ranking pages by keyword.

Good to know

• The numbers are the same ones on our pages, from our own runs, not vendor benchmarks. See how we test.
• Memory figures assume an 8k-token context. We don't measure speed, so the server doesn't either.
• It never returns our hidden test questions or the models' answers to them, only scores.
• We don't log what you ask. Cloudflare, which hosts the site, keeps standard request logs.
• Prefer raw data? The full leaderboard is also available as JSON and CSV.