Ask Claude, ChatGPT, Grok or any MCP app for GPU and AI token prices. Free, read-only, no sign-in.
https://computescreener.com/mcpClaude
- Customize › Connectors › Add custom connector.
- Paste the URL, choose no sign-in, then Add.
ChatGPT
- Plugins (chatgpt.com/plugins) › + › Add custom MCP server.
- Paste the URL, choose no authentication, then create it.
Grok
- grok.com/connectors › New Connector › Custom.
- Paste the URL.
DeepSeek
DeepSeek's chat app has no custom connectors yet. DeepSeek Harness can use the URL as a streamable-http MCP server.
Other apps
Any app that supports remote MCP servers, for example Claude Code: claude mcp add --transport http computescreener https://computescreener.com/mcp
Tools
- GPU rental prices (
get_gpu_prices) — USD / GPU-hour with 1D, 1W and 1M changes. Name one GPU, such as H100 or B200, to also get each provider's average price and the exchanges' prices for contracts on that GPU. - AI token prices (
get_token_prices) — USD / 1M tokens per company. Name one company to also get its models' usage and prices. - Price history (
get_price_history) — any GPU or company, 1 to 1000 days back (30 by default), daily (GPU weekly before 2026-10-23).
All three only read the published prices. Nothing is changed, and nothing about your requests is kept. The answers are CC BY 4.0: reuse them, including commercially, with credit to Compute Screener.
Try asking
- What does it cost to rent an H100 per hour, and how has that changed this month?
- Compare the rental prices of H200, B200 and B300.
- Where do H100 rental contracts trade, and at what price?
- Which AI company's tokens cost the most, and how have prices moved this week?
- Show OpenAI's token price over the last 90 days.
Troubleshooting
- No tools appear: check the URL is exactly the one above and that no sign-in is chosen, then remove the connector and add it again.
- "Not listed yet": a GPU or company is listed once it has a price and a month of history.
- "Prices are unavailable right now": try again in a minute; this website shows the same figures.
- "Too many requests": one network address can send up to 150 requests in 10 seconds; wait 10 seconds.
- Anything else: Support.
Methodology · Support · Privacy · Terms