Getting started / MCP
Your first run starts
with an estimate.
Connect Claude Code or a compatible remote MCP client. Get a price for the work before you add credit for a run.
Give the job a home.
Create or sign in to your Symbioza account using Google, GitHub or email. Your connected agent works with the jobs belonging to that account.
You can connect and request an estimate before adding credit. Paid submission requires a prepaid balance.
Ready when: you can open your account.
One connector URL.
Add a custom remote connector in your MCP client and complete the sign-in it opens. The client must support remote HTTP MCP and authentication.
Using Claude Code?
Claude Code: install the plugin — claude plugin marketplace add symbioza/claude-plugin then claude plugin install symbioza@symbioza. Other MCP clients: add the connector URL.
The Claude Code plugin, ChatGPT and other clients: Integrations →
Check the tool list
describeDataset, estimateExecution, submitJob, getStatus, listJobs, getArtifact and cancelJob. Read the submission details in the Agent guide for request IDs, budget and runtime limits.
Ready when: you ask 'Use Symbioza to list my jobs.' and get a list — an empty one counts.
Describe the work and the limit.
Have your container image, command, data location and intended outputs ready. Your agent can help prepare a valid spec and identify missing requirements.
Use one of the supported images; any other is refused, with the current list, before anything is booked or charged. A supported framework image does not include your code, so have the command fetch your script and anything else it needs.
Ready when: you understand the estimate, total limit and what the job will produce.
Review. Fund. Run.
Review the job and spending limit in your client. If you want to proceed, add prepaid credit and ask your agent to submit the reviewed spec.
Read the Terms before topping up. Top-ups are generally non-refundable, subject to the stated exceptions and mandatory consumer rights.
Ready when: your agent returns an execution identifier. Save it; you can disconnect.
Check the run. Get the files.
Come back through the same account and ask for the job’s status and artifacts. Your command should write the files you want to /workspace/artifacts/.
- Run: did the command finish successfully?
- Delivery: are the expected files available, and do their hashes verify?
- Billing: is the reported charge final or still provisional?
A completed run does not answer all three questions. Download-link expiry also does not mean the stored file has been deleted; see retention details.
Start with the reported reason.
My client says “unauthorized.”
Reconnect and complete authentication. Confirm that you are using the intended account. Client menus and sign-in behavior can differ.
The job is rejected before it starts.
Read the reason: it may concern the spec, balance, spending limit or available compute. Correct the specific issue and request a new estimate if the spec changes. Raising the budget is not the answer to every rejection.
The run failed or lost contact.
Inspect incidents, logs and the charge before resubmitting. A failure can come from the workload or infrastructure. Recovery is conditional; do not assume that all infrastructure failures were already recovered.
The job completed but files are missing.
Check output paths and delivery status. Ask for the artifact manifest again if delivery is still pending. A successful exit code is not proof that every file arrived.
Still stuck? Contact Symbioza with the execution identifier and the error. Leave credentials and signed dataset links out of your message.
Copy the connector URL or a prompt to use in your client.
Bring your next job.
Choose a workload you can define, estimate it, and decide whether to run.