FARPY routes GPU workloads to qualified compute and returns verified artifacts and receipts.
One address for any MCP client. You sign in at api.farpy.com once, after adding it.
At most: moves money, runs code or commands, deletes data, or sends messages for you.
Odel charges nothing per call. The publisher may charge for its own service.
| Access to | Read | Change |
|---|---|---|
| Mcp | Not stated: mcp:directory | |
From the sign-in at api.farpy.com, as the endpoint publishes it. You approve it there.
Odel could not list its tools without signing in.
Verified GPU compute for humans, applications, and agents.
FARPY accepts bounded GPU workloads, executes them asynchronously, delivers completed artifacts, and issues verifiable execution receipts.
Blender Cycles GPU rendering — $0.01 per completed frame.
Current public contract:
.blendLIVE_PUBLIC_ALPHAAdditional workloads are under development and are not currently public.
Typical lifecycle:
discover → price → submit → JOB-ID → execute → download → verify receipt
A local stdio MCP server for Cursor, Claude Desktop, Glama, and other MCP clients lives in mcp/.
It is a thin public client over FARPY's existing Job API. It does not include production server code and never logs FARPY_AGENT_TOKEN.
cd mcp
npm install
npm run build
npm test
node dist/mcp-server.js
Environment (see .env.example):
FARPY_AGENT_TOKEN — Job API farpy_agent_ key. Required for every tool. The server fails closed if it is missing.FARPY_API_BASE — API origin. Default: https://api.farpy.comExample local client config:
{
"mcpServers": {
"farpy": {
"command": "node",
"args": ["/absolute/path/to/farpy-public/mcp/dist/mcp-server.js"],
"env": {
"FARPY_AGENT_TOKEN": "farpy_agent_..."
}
}
}
}
Public MCP V1 tools: inspect, quote, start, status, download, receipt. inspect and quote do not spend. start is the spend boundary and requires legal_acceptance=FARPY_LEGAL_V1.
FARPY is currently a public-alpha service.
Public capabilities, pricing, and limits are authoritative. Historical research, internal tests, and experimental workloads do not imply current public availability.