Match a workload to the cheapest GPU

The routing DECISION: describe a job (a model, size, or GPU need) and get the ranked, reasoned recommendation for the cheapest place to run it across the live m

liveverifiedunknown~5s typical

mcpprobedfastgpu

$0.0020

per call

plus $0.001 platform fee per run

Open MCP server; Cracked bills routing only

Provider Upstream docs

$Set up https://cracked-ai-pearl.vercel.app/SKILL.md, then use Cracked to run mcp-co-fastgpu-fastgpu /match_workload for me.

Input

Field	Type	Description
query	string	Plain-language job, e.g. "cheapest to serve Llama 3 70B" or "2x H100 for fine-tuning". Pro
model	string	Open model name to size against, e.g. "Llama 3 70B", "Qwen 72B", "Mixtral".
params_b	number	Model size in billions of parameters when no exact model is named.
vram_gb	integer	Rough VRAM the job needs, in GB, if you already know it.
task	string (inference \| finetune-lora \| finetune-full \| generate \| transcribe \| embed)	What the job does.
precision	string (fp16 \| int8 \| int4)	Numeric precision to size the model at.
gpu_count	integer	Force a specific GPU count instead of letting the engine size it.
budget_usd_hr	number	Only recommend configs at or under this hourly budget.
region	string (US \| EU \| ASIA)	Restrict to a data-residency region.
spot	string (true \| false)	Set true to include interruptible spot capacity for a cheaper rate.
reserved	string (true \| false)	Set true to include reserved / committed-term capacity for a lower rate.

Call it

curl

curl https://cracked-ai-pearl.vercel.app/v1/run \
  -H "Authorization: Bearer ck_live_..." -H "content-type: application/json" \
  -d '{"provider":"mcp-co-fastgpu-fastgpu","endpoint":"/match_workload","input":{}}'

cli

npx cracked-ai run -p mcp-co-fastgpu-fastgpu -e /match_workload -i '{}'

mcp

run_tool({ provider: "mcp-co-fastgpu-fastgpu", endpoint: "/match_workload", input: {} })

Try it

Runs against your signed-in workspace balance. Sign in if you have not.