compute.pangle.online — GPU rental market
Live GPU rental market: 2,500+ offers across a dozen provider feeds. History, watches, limit orders.
Install
Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).
claude mcp add --transport http compute-pangle-online-gpu-rental-market 'https://compute.pangle.online/mcp/'
{
"mcpServers": {
"compute-pangle-online-gpu-rental-market": {
"type": "http",
"url": "https://compute.pangle.online/mcp/"
}
}
}
Documentation
From the project's README on GitHub, refreshed weekly. Links open on GitHub.
Contents
compute.pangle.online — GPU compute meta-aggregator
One process: provider pollers + SQLite store + ranked REST API + MCP server + the site.
Live at https://compute.pangle.online · API /api/offers, /api/history, /api/idle-history, /api/watches · MCP /mcp/ (streamable HTTP).
Run anywhere (the VPS lift)
docker compose up # or:
python3 -m venv venv && venv/bin/pip install -r requirements.txt
venv/bin/uvicorn app:app --port 8956
No external services, no keys required (keys in .env deepen Vast/RunPod results; DataCrunch and Akash need none).
The fee levers — there are two, and confusing them corrupts the data
FEE_BPS (core/economics.py) marks up displayed prices, and it is 0, permanently.
The book, the tape, the index and the per-chip prices are published as THE MARKET's numbers —
check any of them against the provider's own page. A markup here would quietly make every
published price ours instead of theirs, so this lever stays off.
EXEC_FEE_BPS is the platform fee, charged on what we settle, never on what the provider
charges: 2.5%, one cent minimum, capped at $5 per lease per UTC day (core/meter.py).
Renters place on their own provider key and pay the provider directly; this is the only thing
we take, and it is disclosed in the quote and the receipt before anything is placed.
REF_VAST / REF_RUNPOD referral slots ride the same seam.
Layout
providers/ one drop-in module per marketplace · core/ schema, rank, economics,
stats · db.py SQLite (WAL) with provider_health honesty table · app.py FastAPI +
scheduler + MCP mount · web/ the hand-written site (see DESIGN-DIRECTION.md).
Tests: venv/bin/python -m tests.test_pipeline
Generated pages
web/gpu/, web/vs/ and web/sitemap.xml are built artifacts — regenerate with
venv/bin/python gen_gpu_pages.py (the deployment runs it hourly).
Fonts
Martian Mono ships in web/fonts/ (SIL OFL). The display serif (Erode, by
Indian Type Foundry via Fontshare) is not redistributed here — its license
doesn't allow it; fetch it from https://www.fontshare.com/fonts/erode into
web/fonts/ as erode-400.woff2 / erode-600.woff2, or let the stack fall
back to Georgia.
License
AGPL-3.0 — see LICENSE. Run it, fork it, learn from it; if you serve a
modified version, share your changes the same way.
Frequently asked questions
What is compute.pangle.online — GPU rental market?
Live GPU rental market: 2,500+ offers across a dozen provider feeds. History, watches, limit orders.
How do I install compute.pangle.online — GPU rental market?
It runs as a remote server. In Claude Code: claude mcp add --transport http compute-pangle-online-gpu-rental-market 'https://compute.pangle.online/mcp/'. For other clients, paste the JSON configuration from this page into their MCP settings.
Does compute.pangle.online — GPU rental market need an API key?
Its registry entry lists no required key or secret. Some features may still need an account with the service it connects to.
Is compute.pangle.online — GPU rental market open source?
Yes. Its code is on GitHub under the AGPL-3.0 license.
Who maintains compute.pangle.online — GPU rental market?
It is published by aitools420. AI.tn lists it but does not maintain or endorse it.
From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.