Bring your own method
Use scientific presets, a Python run(ctx) program, or a custom command. Upload your inputs through the REST API and reference them in the job.
RESEARCH · ROBOTICS · AUTOMATION
Your workflow.
Real GPU experiments.
Inspectable results.
Run scientific presets or your own code, distribute parameter sweeps across GPU nodes, and bring calculation reports back into your workflow.
One GPU to several nodes
Run independent parameter sets on 1–4 nodes. Dedicated clusters can be attached with the local SSH connector. A single solver needs its own distributed implementation.
Follow the calculation
Download a report for each run: recorded methods, parameters, numerical checks, data series, file hashes and execution details. Custom programs can record equations, inputs, results and explanations.
N8N
Quote → run → inspect → collect
A downloadable workflow using standard n8n nodes. It starts in quote-only mode, checks the rental budget and quote expiry, retries launches with the same idempotency key, polls progress and requests cancellation if its polling limit is reached.
- Create your EMT account, accept the service terms and create an API key with
read,rentandmanagescopes. - Import the workflow JSON into n8n. Create an HTTP Header Auth credential: name
Authorization, valueBearer YOUR_EMT_API_KEY. Select it on every HTTP Request node. - Edit Configuration: experiment, nodes, duration and maximum rental budget. Run once in quote-only mode. To rent, set a persistent unique idempotency key and explicitly enable
allowPaidRun. - After a lost launch response, inspect your EMT jobs or retry that node with its saved input. To resume monitoring, set
existingJob. Keep the same payload and key for a retry.
Prices use USD micro-units: 1 USD = 1,000,000. The template caps GPU rental cost; separately retained storage is additional. It does not enable automatic paid continuation. A pending release still needs attention.
MODEL CONTEXT PROTOCOL
Give your tools access to compute
Connect a client that supports Streamable HTTP and custom authorization headers. Public tools browse GPU availability and scientific methods. Private tools inspect your jobs, quote GPU nodes or clusters, launch authorized experiments and retrieve report links.
Transport: Streamable HTTP
https://emtcompute.com/mcp
Header
Authorization: Bearer YOUR_EMT_API_KEY
Use a read key for inspection; add rent for quotes and paid launches, and manage for cancellation or cluster attachment. Paid tools require confirm_paid_rental: true. Configure a spending limit on the key. Never put credentials in a prompt or URL.
The server exposes 17 tools. It uses scoped API keys; OAuth-only clients need a compatible MCP bridge or the local Python stdio adapter, which exposes the infrastructure toolset.
CUSTOM CODE & CLUSTERS
Keep your calculations explainable
ctx.explain(
"Kinetic energy",
equation="E = 0.5 × m × v²",
inputs={"mass_kg": 2, "velocity_m_s": 3},
result=9, unit="J",
explanation="Calculated from the measured mass and speed."
)The report records what the program supplies and what the runtime can observe. Numerical checks, source hashes and device information help you inspect the result; they are not an independent proof of arbitrary code. Unsaved intermediate tensors and GPU instructions cannot be reconstructed.
For a dedicated cluster, attach the experiment in the console or through MCP, then run emt_cluster.py locally with your SSH identity. Keep emt_compute.py in the same folder. The entire attached cluster is released when the experiment completes or is cancelled. Preparation time is billed.
export EMT_API_KEY="YOUR_EMT_API_KEY" python3 emt_cluster.py --job JOB_UUID --identity /path/to/private_key
Available nodes and cluster fabrics vary. Check the live catalog and service status before quoting. Fabric declarations and software tests do not establish distributed solver performance.