What the tools answer

get-gpu-price takes one chip, one provider, one region and one price type. It answers with the price per GPU-hour now, its last change, the source URL and when the price was read.

find-cheapest-gpu-hour takes one chip, and optionally one region. It answers with the cheapest current list price per GPU-hour across providers.

Three more tools sit beside them on the same MCP server. list-price-changes lists the changes since a date for a chip or a provider, at most 100 rows. get-compute-contract returns one compute derivative contract with its index and venue. list-regulator-actions lists regulator actions and due dates on traded compute. Each of the five also answers as a JSON endpoint.

Who it is for

Agents that need a GPU price they can cite: to estimate what a job will cost, to compare providers, or to choose a region. Every answer carries its source and the day it was read, so the agent can show where the number came from.

What it costs

On the MCP server, the first 50 calls per caller each UTC day cost nothing. After that each call is $0.01 in USDC on Base, paid with x402 through DAP Agent Payments, with no account and no key.

The JSON API charges $0.01 from the first call. A lookup that matches nothing answers 404 and is not charged. Institutions can use one DAP licence key instead, billed by invoice.

Every DAP tool for agents, with its price

Example: one price, in one region

MCP request (tools/call)
POST https://dap.solutions/compute/mcp
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get-gpu-price","arguments":{"chip":"H100","provider":"AWS","region":"us-east-1"}}}
The same question to the JSON API
GET https://dap.solutions/compute/api/gpu-price?chip=H100&provider=AWS&region=us-east-1

The answer on 4 November 2026 was one line:

Instance
p5.48xlarge, 8 GPUs
Region
US East (N. Virginia)
Price type
On-demand
Price per GPU-hour
$6.88, derived from $55.04 per instance-hour as listed
Price since
2026-11-01 (AWS publishes one price list a month)
Last change
2025-06-01, down 44 percent, from $12.29 to $6.88 per GPU-hour
Source
https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/index.json
Read
2026-11-04

Example: the cheapest hour

MCP request (tools/call)
POST https://dap.solutions/compute/mcp
{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"find-cheapest-gpu-hour","arguments":{"chip":"H100"}}}

The answer on 4 November 2026: the cheapest on-demand H100 was $6.88 per GPU-hour, on AWS p5.48xlarge, and three regions tied at that price: us-east-1 (US East, N. Virginia), us-east-2 (US East, Ohio) and us-west-2 (US West, Oregon). The tool compared 34 lines from AWS and Azure.

These are list prices only. Reserved terms and private contracts can be lower.

How to read the numbers

The price per GPU-hour is derived. It is the provider's listed instance price divided by the number of GPUs on the provider's own specification page: $55.04 per instance-hour for 8 GPUs is $6.88 per GPU-hour.

Each provider's price is its own. The tracker never averages or blends prices across providers.

Both answers are from 4 November 2026, snapshot compute-2026-W45, with the AWS Price List read on 2026-11-04. Prices move: the same call on another day can give a different answer.