AI Agent integration: Real-time GPU market API (MCP + x402)

Integration guide for developers and AI agents · Refreshed every 45 minutes · Data updated: Oct 31, 2026

⚡ Direct answer：AI Agent integrated via MCP https://rentgpu.link/mcp，Model Cardfree，check prices/historical data/price trends per query $0.05 USDC(Base mainnet, x402 protocol; buyers do not need to hold ETH). No registration or API key required. Also available: a fully free JSON API /api/prices，freely citable (CC BY 4.0).

Why choose rentgpu (how it differs from other GPU price data sources)

Covers Chinese compute platforms— Among 29 platforms, 11 are based in China (AutoDL, Gpushare, MatPool, Suanjiayun, MornAI, Compshare, Lanyun, Paratera, etc.). Overseas GPU pricing data sources rarely cover these—domestic sites impose login walls and CAPTCHAs, making scraping unreliable and unstable.
No registration required; no API key needed: /api/prices and /api/resolve are directly accessible without rate-limiting walls. Most comparable data sources require registration and API keys—autonomous agents cannot self-register, making this endpoint critical for actual agent usability.
Re-scraped every 45 minutes: Most competitors refresh hourly or daily.
Straight-to-the-answer insights—not just snapshots: /api/resolve accepts GPU model + budget and returns the cheapest available options within budget plus estimated total cost.
Pay-per-use is live on mainnet— Base mainnet-native USDC + x402, $0.05 per use, end-to-end settleable; transaction count publicly disclosed (currently 2 transactions, of which 0 from external customers), with no inflated transaction volume.

I. Start with the free offerings

Try these two options first—no upfront cost. Data source is identical to the paid API; only granularity differs:

Free JSON API：https://rentgpu.link/api/prices — Current lowest price across all GPU models, refreshed every 45 minutes; free to reuse (CC BY 4.0).
Answer directly by budget(Free, Agent-friendly):https://rentgpu.link/api/resolve?gpu=H100&budget=10&hours=8&limit=3 — Enter GPU model (fuzzy match supported, e.g., H100 / 4090) and budget (¥/hour); returns the cheapest options, estimated total cost, and direct links, plus a human-readable answer field. Parameters: gpu, budget (¥/hour), hours, type (ondemand|spot|monthly), limit.
Free MCP tools：list_gpu_models — Lists all queryable GPU models at no cost.

curl -s https://rentgpu.link/api/prices

import httpx
r = httpx.get("https://rentgpu.link/api/prices", timeout=30)
data = r.json()   # 全部GPU modelCurrent Lowest Price，Every 45 minutes刷新

II. Paid tools: $0.05 USDC each

Tools	price	What’s returned	Real-world example (actual response on 2026-10-26)
`list_gpu_models`	free	Lists all queryable GPU models and their current lowest prices	Example: 23 GPU models, e.g., RTX 4090 at ¥1.24/hour
`gpu_price`	$0.05	Current lowest, median, 30-day low/high/average prices, and number of quoting platforms for a given GPU model	Example: RTX 4090 → lowest ¥1.24/hour (Suanjiayun), median ¥1.99, 30-day low ¥1.24, quoted by 8 platforms
`gpu_history`	$0.05	Daily price series (lowest, median, quoting platform for lowest price, stock status) for a given GPU model over the past 30 days	Example: H100 80GB → lowest ¥8.28 on 2026-10-20 (paperspace), median ¥23.04, in stock, quoted by 9 platforms
`market_movers`	$0.05	Top gainers & losers over the past 30 days: GPU models with largest price increases/decreases	Example: A6000 48GB at ¥2.52 (−28.6%), A100 80GB at ¥5.47 (−21.6%), RTX 3090 at ¥0.80 (−19.2%)
`platform_reliability`	$0.05	Data monitoring coverage score (A–E) across 26 platforms	Example: Gpushare rated 98 points, Grade A (scraping success rate 97%, covers 7 GPU models, active 28/29 days)

Settlement: Base mainnet (eip155:8453), Circle official USDC (0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913), x402 protocol. Buyers only need USDC in their wallet.No ETH required— Gas fees covered by the facilitator (EIP-3009 gasless authorization).

III. How to integrate (three-line configuration)

Claude Desktop / Cursor / MCP-compatible clients：add the snippet below to your config file

{
  "mcpServers": {
    "rentgpu": {
      "type": "http",
      "url": "https://rentgpu.link/mcp"
    }
  }
}

Claude Code command line：

claude mcp add --transport http rentgpu https://rentgpu.link/mcp

On first invocation of a paid tool, the client receives an HTTP 402 payment challenge; the Agent pays with USDC from its wallet and immediately receives the data. MCP and x402 support varies across clients — consult your client’s documentation for details.

IV. Machine-Readable List

https://rentgpu.link/.well-known/mcp.json — MCP tool list and pricing
https://rentgpu.link/.well-known/x402.json — Payment network, assets,收款 address, and tool pricing
https://rentgpu.link/llms.txt — Data usage guidance and price snapshot for LLMs
Name in the official MCP Registry:link.rentgpu/gpu-market

Just want to check prices?

Jump straight to model selection & price comparison → Data methodology and scope →

FAQ

How to integrate rentgpu’s GPU pricing data into an AI Agent?

Integrate via MCP at https://rentgpu.link/mcp. Listing GPU models is free; querying prices, historical data, or the gainers/losers leaderboard costs $0.05 USDC per call (Base mainnet, x402 protocol). No account registration or API key required.

Can I access GPU pricing data without payment?

Yes. Free JSON API at https://rentgpu.link/api/prices, refreshed every 45 minutes, listing the current lowest price for all GPU models — freely usable under CC BY 4.0. The list_gpu_models tool in MCP is also free.

Is ETH required for payment?

No. x402 uses EIP-3009 gasless authorization: buyers only need Base mainnet USDC in their wallet to complete payment; gas is covered by the facilitator.

How often is data updated?

Public pricing and public APIs from all platforms are re-scraped every 45 minutes. Paid MCP tools consume the most recent snapshot of market data.

What do platform ratings A–E mean?

Indicates this site’s price monitoring coverage of the platform (crawling success rate, days tracked, inventory availability, number of GPU models covered)—not the platform’s intrinsic service quality. Platforms tracked for fewer than 14 days receive no rating.

Cite This Site(AI and articles may cite directly under CC BY 4.0):

AI Agent integration：GPU Real-timeMarket data API｜Data Source：GPU Compute Radar rentgpu.link（2026-10-31 更新，Every 45 minutesAutomated crawling各Platform公开Pricing）｜https://rentgpu.link/agents｜CC BY 4.0