Skip to content
modelread

Loading devices…

Remembered on this device. Multi-GPU: search “2x”, “8x”, “node”.

Quickstart

modelread is a public, read-only service. There is no account and no API key. Pick one way in:

1. As an MCP server (recommended for agents)

Endpoint: https://modelread.com/mcp (streamable HTTP).

claude mcp add --transport http --scope user modelread https://modelread.com/mcp

Other clients (Cursor, VS Code, Codex, Gemini CLI, Windsurf, Zed, LM Studio, Claude.ai) have one-click buttons or config snippets on Connect. Then just ask your agent things like “will Qwen/Qwen3.6-27B run on my RTX 4090 at 64K context?”

2. As an HTTP API

curl "https://modelread.com/api/estimate?repo=Qwen/Qwen3.6-27B&hw=rtx4090&ctx=32768"

See HTTP API for every endpoint.

3. In the browser

Ask on the home page, open any model at /m/{org}/{repo}, or any device at /hw/{id}. Your device choice is remembered locally and used as the default.

4. As an agent skill

mkdir -p ~/.claude/skills/modelread && curl -fsSL https://modelread.com/skill.md -o ~/.claude/skills/modelread/SKILL.md

The skill tells Claude Code when to check fit before downloading, quantizing or serving a model.

Askwhat should I run?
Browsebest models per device
Devicesall hardware
Connect your agentMCP · API
How the numbers workcalibration