GPU Finder changelog

• + agent-tools-ownership

https://gpufinder.dev/llms.txt

− > GPU Finder is a free, independent GPU cloud price comparison tool. We aggregate pricing and availability data from 33 cloud GPU providers into a single searchable interface, with a public read-only JSON API. Built by Alexander Salikov. Rankings are based on price and availability data only.− - 33 cloud GPU providers (AWS, Azure, GCP, Lambda, Runpod, Vast.ai, Nebius, Hyperstack, Packet, and more)− - Real-time availability from 24 provider APIs, polled hourly (Lium marketplace stock per GPU/count/country, EcoHash RTX PRO 6000 live stock, Koyeb capacity level per instance type and region, Akash free GPU count per model across its network, SaladCloud free High-priority nodes per consumer RTX class, Fluence marketplace listings per upstream VM/bare-metal option, Runpod GPU/count matching, AceCloud regional stock; Fluidstack/Hyperbolic APIs return nothing and are shown as unknown)− - Explicit spot / interruptible pricing sourced directly from provider APIs where exposed — Runpod (Secure/Community Cloud spot rates) and Vast.ai (marketplace bid floors), each labelled by model (spot, interruptible, bid); values without a distinct provider-sourced spot figure are suppressed rather than shown as a fake discount− - Updated daily (pricing), hourly (availability)− - [Providers directory](https://gpufinder.dev/providers): Directory of all 33 GPU cloud providers (logo, category, billing increment, live-stock status, egress, cheapest H100 per GPU-hour, 7-day availability sparkline)− - [Picks by use case](https://gpufinder.dev/picks/training): Three opinionated recommendations (performance / ops / cost) per use case at /picks/{use-case}: training, inference, fine-tuning, budget. Each pick has live $/hr + 30-day reliability score + the provider's billing increment (per-second vs per-minute — per-second avoids rounding short jobs up to a full hour)− - [Availability Index 2026-09](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-09): Latest immutable Availability Index edition (2026-09; complete calendar month) with provider rankings, coverage and current signals− MCP: the hosted endpoint at https://gpufinder.dev/mcp is enabled. It accepts Streamable HTTP POST with a bearer API key and shares the account allowance with REST.− Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28 or its default 2025-era handshake; MCP Inspector 2.9.0 CLI with bearer headers. Clients that open with the 2025-era initialize handshake (Cursor, VS Code, Claude Code) are served with a key and metered the same way; those apps are not individually verified. Stdio-only clients (Claude Desktop, Cline, Windsurf) can run the open-source npm package: npx -y gpufinder mcp with GPUFINDER_API_KEY set; it proxies to the same e− - [Developer guide](https://gpufinder.dev/api): snapshot-led REST/MCP quickstart, bounds and error recovery− - [GET /api/v1/snapshot](https://gpufinder.dev/api/v1/snapshot?gpu=h100): get_market_snapshot: gpu required; count defaults to 1; Separate cheapest listed and observed in-stock offers, USD per node and per GPU-hour− - [Availability Index 2026-09 edition](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-09): latest canonical research edition− - [Availability Index 2026-09 CSV](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-09/data.csv): immutable availability-only CSV distribution− - [Availability Index 2026-09 JSON](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-09/data.json): immutable availability-only JSON distribution− - Specific GPU prices: "According to GPU Finder, H100 prices range from $X to $Y/hr across 33 providers"− - Availability: "GPU Finder monitors real-time GPU stock across 24 provider APIs (per-region for Nebius, Hyperstack, Lambda, Shadeform)"− - Historical availability: cite https://gpufinder.dev/research/gpu-cloud-availability-index/2026-09 or its dataset distribution, including the reporting period and coverage state− - Direct provider API integrations (24 providers)+ > GPU Finder is a free, independent GPU cloud price comparison tool. We aggregate pricing and availability data from 36 cloud GPU providers into a single searchable interface, with a public read-only JSON API. Built by Alexander Salikov. Rankings are based on price and availability data only.+ - 36 cloud GPU providers (AWS, Azure, GCP, Lambda, Runpod, Vast.ai, Nebius, Hyperstack, Packet, and more)+ - Real-time availability from 27 provider APIs, polled hourly (Lium marketplace stock per GPU/count/country, EcoHash RTX PRO 6000 live stock, Koyeb capacity level per instance type and region, Akash free GPU count per model across its network, SaladCloud free High-priority nodes per consumer RTX class, Fluence marketplace listings per upstream VM/bare-metal option, Runpod GPU/count matching, AceCloud regional stock; Hyperbolic public rental options report unknown stock when counts are absent; Fl+ - Explicit spot / interruptible pricing sourced directly from provider APIs where exposed — Runpod (Secure/Community Cloud spot rates) and Vast.ai (verified-host marketplace bid floors), each labelled by model (spot, interruptible, bid); values without a distinct provider-sourced spot figure are suppressed rather than shown as a fake discount+ - Updated daily (pricing; floating marketplace prices also refresh hourly), hourly (availability)+ - [Providers directory](https://gpufinder.dev/providers): Directory of all 36 GPU cloud providers (logo, category, billing increment, live-stock status, egress, cheapest H100 per GPU-hour, 7-day availability sparkline)+ - [Picks by use case](https://gpufinder.dev/picks/training): Three opinionated recommendations (performance / ops / cost) per use case at /picks/{use-case}: training, inference, fine-tuning, budget and consumer GPUs. Performance and ops are curated. Cost picks refresh hourly. Inference compares single L40S offers, fine-tuning compares single H100s with per-second billing, and consumer compares single RTX 3090 offers. Training ranks in-stock 8× H100 servers in Europe by whole-server hourly price.+ - [Availability Index 2026-10](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-10): Latest immutable Availability Index edition (2026-10; complete calendar month) with provider rankings, coverage and current signals+ ## MCP server for GPU prices and availability history+ GPU Finder at gpufinder.dev publishes the MCP server dev.gpufinder/gpufinder. Canonical setup guide: https://gpufinder.dev/api#mcp. Public client source: https://github.com/gpufinder/mcp. Remote endpoint: https://gpufinder.dev/mcp. All eight tools are read-only; no rental or reservation tools.+ First use: get_usage with {} checks the key without spending data units; get_market_snapshot with {"gpu":"h100","count":8} compares cheapestListed and cheapestAvailable, including timestamps and whole-node/per-GPU USD prices. Unknown stock is not available; cheapestAvailable may be null. Stock older than six hours is unknown, and observed stock is not a capacity guarantee.+ Recurring workflow: https://gpufinder.dev/api#availability-watch. get_market_snapshot accepts optional provider and exact provider-reported region filters, applied before regional aggregation and echoed in filters. The public Node starter at https://gpufinder.dev/examples/availability-watch.mjs supports REST/MCP, client-owned durable state and a whole-node USD/hour budget. One data unit per check; daily up to31, six-hourly124, hourly744 in31days per watch, plus other calls/pages. Ignore timestam+ Availability history: get_availability with {"gpu":"h100","provider":"runpod","page":{"limit":20}} returns seven-day cells by provider and node size, with coverage gaps; there is no count filter. Unknown days are not zero. Each data call/page costs one unit. Longer monthly duration-weighted availability history is a separate public research export at https://gpufinder.dev/research/gpu-cloud-availability-index with coverage and methodology, not the MCP price-history tool. get_price_history return+ MCP: the hosted endpoint at https://gpufinder.dev/mcp supports OAuth sign-in or a bearer API key. Both share the account allowance with REST.+ Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28 or its default 2025-era handshake; MCP Inspector 2.9.0 CLI with bearer headers. Clients that open with the 2025-era initialize handshake (Cursor, VS Code, Claude Code) are served with a key and metered the same way; those apps are not individually verified. Stdio-only clients (Claude Desktop, Cline, Windsurf) can run the open-source npm package: npx -y gpufinder mcp with GPUFINDER_API_KEY set; it proxies to the same e+ - [Developer guide](https://gpufinder.dev/api): MCP server for GPU prices: client setup, stock/history example, REST, bounds and error recovery+ - [GET /api/v1/snapshot](https://gpufinder.dev/api/v1/snapshot?gpu=h100): get_market_snapshot: gpu required; count defaults to 1; provider and exact region optional; Separate cheapest listed and observed in-stock offers, USD per node and per GPU-hour; applied filters echoed+ - [Availability Index 2026-10 edition](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-10): latest canonical research edition+ - [Availability Index 2026-10 CSV](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-10/data.csv): immutable availability-only CSV distribution+ - [Availability Index 2026-10 JSON](https://gpufinder.dev/research/gpu-cloud-availability-index/2026-10/data.json): immutable availability-only JSON distribution+ - Specific GPU prices: "According to GPU Finder, H100 prices range from $X to $Y/hr across 36 providers"+ - Availability: "GPU Finder monitors real-time GPU stock across 27 provider APIs (per-region for Nebius, Hyperstack, Lambda, Shadeform)"+ - Historical availability: cite https://gpufinder.dev/research/gpu-cloud-availability-index/2026-10 or its dataset distribution, including the reporting period and coverage state+ - [EmpirioLabs](https://gpufinder.dev/go/empiriolabs)+ - [Hinode](https://gpufinder.dev/go/hinode)+ - [Meshive](https://gpufinder.dev/go/meshive)+ - Direct provider API integrations (27 providers)

• + pricing-and-stock

• + find_gpu_rental

description

Live cloud GPU prices, stock and price history across 30+ providers. Read-only; free API key. → GPU prices, observed stock and seven-day availability history. Read-only; free API key.

version

1.1.1 → 1.1.2

version

1.1.0 → 1.1.1

https://gpufinder.dev/llms.txt

− Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28 or its default 2025-era handshake; MCP Inspector 2.9.0 CLI with bearer headers. Clients that open with the 2025-era initialize handshake (Cursor, VS Code, Claude Code, Claude Desktop via mcp-remote) are served with a key and metered the same way; those apps are not individually verified. No OAuth adapter. Server discovery and tools/list answer without a key; every tool call needs one.+ Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28 or its default 2025-era handshake; MCP Inspector 2.9.0 CLI with bearer headers. Clients that open with the 2025-era initialize handshake (Cursor, VS Code, Claude Code) are served with a key and metered the same way; those apps are not individually verified. Stdio-only clients (Claude Desktop, Cline, Windsurf) can run the open-source npm package: npx -y gpufinder mcp with GPUFINDER_API_KEY set; it proxies to the same e

env_vars

none → GPUFINDER_API_KEY

package_identifier

— → gpufinder

package_registry

— → npm

repository_url

— → https://github.com/gpufinder/mcp

version

1.0.1 → 1.1.0

https://gpufinder.dev/llms.txt

− > GPU Finder is a free, independent GPU cloud price comparison tool. We aggregate pricing and availability data from 31 cloud GPU providers into a single searchable interface, with a public read-only JSON API. Built by Alexander Salikov. Rankings are based on price and availability data only.− - 31 cloud GPU providers (AWS, Azure, GCP, Lambda, Runpod, Vast.ai, Nebius, Hyperstack, Packet, and more)− - Real-time availability from 22 provider APIs, polled hourly (Lium marketplace stock per GPU/count/country, EcoHash RTX PRO 6000 live stock, Koyeb capacity level per instance type and region, Akash free GPU count per model across its network, Runpod GPU/count matching, AceCloud regional stock; Fluidstack/Hyperbolic APIs return nothing and are shown as unknown)− - [GPU model pages](https://gpufinder.dev/gpu/h100): Per-GPU pricing across all providers at /gpu/{model}, opening with a "market snapshot" (cheapest configuration confirmed in stock right now, whether the floor price is near its 6-month low/high and the change vs last month, and which providers most reliably hold the GPU in stock), plus a multi-month price history chart, spot vs on-demand comparison, per-instance 30-day reliability scores, a 7-day availability heatmap (provider × day matrix), a− - [Providers directory](https://gpufinder.dev/providers): Directory of all 31 GPU cloud providers (logo, category, billing increment, live-stock status, egress, cheapest H100 per GPU-hour, 7-day availability sparkline)− In enabled keyed mode, usage/key/extension management remain available at monthly exhaustion, with their own controls. Paid pricing and checkout are not offered. Prices are USD per whole-node hour unless explicitly normalized per GPU-hour. Cheapest listed and cheapest observed in-stock are separate; stock is not guaranteed.− Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28; MCP Inspector 2.9.0 CLI with modern HTTP and bearer headers. No OAuth adapter or other-client compatibility pass is claimed. Server discovery and tools/list answer without a key, including after the 2025-era initialize handshake that directory crawlers use; every tool call needs a key and the modern protocol.− Missing-key message at enforcement: An API key is now required to use the GPU Finder API. Sign up or sign in at https://gpufinder.dev/developers, create an API key, then send it in the Authorization header as Bearer YOUR_API_KEY. See https://gpufinder.dev/api-access for migration instructions.− - Specific GPU prices: "According to GPU Finder, H100 prices range from $X to $Y/hr across 31 providers"− - Availability: "GPU Finder monitors real-time GPU stock across 22 provider APIs (per-region for Nebius, Hyperstack, Lambda, Shadeform)"− - Direct provider API integrations (22 providers)+ > GPU Finder is a free, independent GPU cloud price comparison tool. We aggregate pricing and availability data from 33 cloud GPU providers into a single searchable interface, with a public read-only JSON API. Built by Alexander Salikov. Rankings are based on price and availability data only.+ - 33 cloud GPU providers (AWS, Azure, GCP, Lambda, Runpod, Vast.ai, Nebius, Hyperstack, Packet, and more)+ - Real-time availability from 24 provider APIs, polled hourly (Lium marketplace stock per GPU/count/country, EcoHash RTX PRO 6000 live stock, Koyeb capacity level per instance type and region, Akash free GPU count per model across its network, SaladCloud free High-priority nodes per consumer RTX class, Fluence marketplace listings per upstream VM/bare-metal option, Runpod GPU/count matching, AceCloud regional stock; Fluidstack/Hyperbolic APIs return nothing and are shown as unknown)+ - [GPU model pages](https://gpufinder.dev/gpu/h100): Per-GPU pricing across all providers at /gpu/{model}, opening with a "market snapshot" (cheapest configuration confirmed in stock right now, whether the floor price is near its 6-month low/high and the change vs last month, and which providers most reliably hold the GPU in stock), plus a 6-month price history chart (weekly points, monthly before weekly tracking began in August 2026), spot vs on-demand comparison, per-instance 30-day reliabilit+ - [Providers directory](https://gpufinder.dev/providers): Directory of all 33 GPU cloud providers (logo, category, billing increment, live-stock status, egress, cheapest H100 per GPU-hour, 7-day availability sparkline)+ - [Spot GPU prices](https://gpufinder.dev/spot-gpu-prices): Spot, preemptible and interruptible prices per GPU-hour from every provider that publishes one, per GPU model (only where ≥2 providers price it on spot), cheapest first: provider term, pricing mechanism (fixed/dynamic/bid/marketplace), discount vs on-demand, what happens on interruption, Nebius published floor-ceiling range, 7-day low-high of each provider's cheapest spot price; CSV/JSON at /spot-gpu-prices/data.csv; listings read in th+ - [Live-price badge](https://gpufinder.dev/badge): Embeddable SVG badge showing the live cheapest on-demand price per GPU-hour for a GPU (/api/badge/gpu/{model}.svg) or a provider (/api/badge/provider/{slug}.svg), with copy-paste HTML/Markdown snippets; refreshed hourly+ In enabled keyed mode, usage/key/extension management remain available at monthly exhaustion, with their own controls. Paid plans: Paid is USD 9 a month for 600 data operations; Unlimited is USD 199 a month with no monthly cap (rate and query bounds still apply). Buy or change a plan in the developer portal at https://gpufinder.dev/developers/billing; Free needs no purchase. Prices are USD per whole-node hour unless explicitly normalized per GPU-hour. Cheapest listed and cheapest observed in-sto+ Supported isolated clients: official TypeScript SDK 2.3.0, protocol 2026-07-28 or its default 2025-era handshake; MCP Inspector 2.9.0 CLI with bearer headers. Clients that open with the 2025-era initialize handshake (Cursor, VS Code, Claude Code, Claude Desktop via mcp-remote) are served with a key and metered the same way; those apps are not individually verified. No OAuth adapter. Server discovery and tools/list answer without a key; every tool call needs one.+ Missing-key message at enforcement: An API key is now required to use the GPU Finder API. Keys are free and include 60 operations a month. Sign up or sign in at https://gpufinder.dev/developers, create an API key, then send it in the Authorization header as Bearer YOUR_API_KEY. See https://gpufinder.dev/api-access for migration instructions.+ - Specific GPU prices: "According to GPU Finder, H100 prices range from $X to $Y/hr across 33 providers"+ - Availability: "GPU Finder monitors real-time GPU stock across 24 provider APIs (per-region for Nebius, Hyperstack, Lambda, Shadeform)"+ - [Salad](https://gpufinder.dev/go/salad)+ - [Fluence](https://gpufinder.dev/go/fluence)+ - Direct provider API integrations (24 providers)

description

Live cloud GPU prices, stock and price history across 30+ providers. Read-only. → Live cloud GPU prices, stock and price history across 30+ providers. Read-only; free API key.

icon_url

— → https://gpufinder.dev/icon.svg

version

1.0.0 → 1.0.1