HBM for AI agents

Give your AI agent real GPU memory data.

Endpointhttps://www.hbmrwa.xyz/api/mcp

Streamable HTTP. No key, no sign-in, open to any origin. 120 requests per minute per address.

Connect

Claude Code

Run in your terminal.

Terminal

claude mcp add --transport http hbm https://www.hbmrwa.xyz/api/mcp

Claude Desktop and claude.ai

Customize, Connectors, +, Add custom connector. Paste the URL and choose No sign-in.

URL

https://www.hbmrwa.xyz/api/mcp

Cursor

Add to your project, or to ~/.cursor/mcp.json for every project.

.cursor/mcp.json

{
  "mcpServers": {
    "hbm": {
      "url": "https://www.hbmrwa.xyz/api/mcp"
    }
  }
}

VS Code

Add to your workspace, or run MCP: Add Server, choose HTTP and paste the URL.

.vscode/mcp.json

{
  "servers": {
    "hbm": {
      "type": "http",
      "url": "https://www.hbmrwa.xyz/api/mcp"
    }
  }
}

ChatGPT

Turn on Developer mode in Settings, Security and login. Then open Plugins, press + and paste the URL.

URL

https://www.hbmrwa.xyz/api/mcp

Other clients

Clients that only speak stdio can bridge to the URL.

Terminal

npx -y mcp-remote https://www.hbmrwa.xyz/api/mcp

Tools

estimate_inference(model, gpu, quant?, ctx?, gpus?, vram_gb?)Whether a model fits a GPU, VRAM needed against available, and the decode ceiling in tokens per second.
list_models()The open models and weight formats the estimate knows.
list_gpus(segment?)Every GPU class with memory, measured and published bandwidth, and verified runs.
gpu_class(gpu)One class: measured median read against the published peak, efficiency, runs and median MEM Score.
leaderboard(segment?, limit?)Top GPU classes by median MEM Score.
network_stats()Operators, verified runs, dies, data tested, the latest on-chain proof root and rewards paid today.
verify_run(run_id)A run's result and its Merkle proof, checked against the day's root on Robinhood Chain.

Every tool answers with a one-line summary and the data as JSON. The numbers are the ones behind Inference, Leaderboard and Network; the API docs have the HTTP routes.

Try asking

Which GPU should I buy to run Llama 70B at 30 tok/s?
How does my RTX 5090 compare to its spec?
Verify HBM run <id> on-chain