How Claude started a temporary server, ran an audit, saved the results, and shut it down in nine minutes.
Nine minutes after starting an SEO audit, Claude had produced a set of reports and terminated the cloud machine that ran the job. The results stayed in storage, ready for review. The machine had finished its work.
That is the workflow in our video demonstration: an AI assistant starts temporary compute, runs an audit, saves the output, and cleans up after itself.
Getting there started with a familiar problem. We wanted to check our websites for SEO issues, but had reached the free URL limit in the tool we were using. Could open-source alternatives provide the coverage we needed?
Claude helped answer that question by running them.
Three tools, three machines, actual results
We asked Claude to evaluate three open-source SEO projects. It deployed each project on its own compute node, keeping the environments separate for testing.
That gave us actual crawl results to compare. We worked through what each tool found, what it missed, and how its coverage compared with the Screaming Frog reports we had been using.
For our audit needs, the evaluation led to a combination of two tools. We then asked Claude to turn that combination into a pipeline and a reusable skill.
This is where remote code execution changes what an assistant can contribute. With access to a machine, it can install software, run an experiment, and bring back evidence. The evaluation becomes something you can inspect and discuss together.
Package the useful work into a skill
The skill captures the process so it can be run again against another domain. In the demonstration, we invoke it against a website we own and control.
Claude starts an SEO audit node using noBGP Compute. Once the machine is online, it copies the prepared crawler applications from noBGP cloud storage onto the node.
Keeping those files in storage avoids rebuilding the applications from their GitHub projects on every run. The initial evaluation does the setup work; subsequent audits reuse the prepared files.
Claude uses noBGP's MCP tools to operate the machine and run the pipeline. The same conversation that starts the job can follow its progress and retrieve its output.
The resulting workflow has five steps:
- Start temporary compute. Create a machine for the audit.
- Prepare the environment. Copy the crawler applications from cloud storage.
- Run the audit. Execute the pipeline against the chosen domain.
- Save the results. Write the reports, data, and summary to persistent storage.
- Terminate the machine. Release the compute once the work is complete.
Keep the output after the machine is gone
At the end of the demonstration, a timestamped run folder contains the audit data and summary. The account view confirms that the SEO audit node has terminated after running for nine minutes.
This separation is central to the workflow. Compute is temporary. The results remain available.
We can download the reports, review them ourselves, or give them back to Claude for analysis. The demonstration stops at producing and saving the findings. Asking the assistant to investigate or fix the website is a follow-up task.
Saving each run also gives us a basis for future comparisons. We can revisit earlier findings and investigate what has changed without keeping the audit machine running between scans.
Make the job recurring
Once the skill worked, we configured a routine to invoke it automatically. The video describes the intended weekly workflow: start compute, run the scan, and store the results for analysis.
The recording shows one completed run of that skill. The reusable process and scheduled trigger are what make it suitable for ongoing monitoring.
A weekly audit only needs its execution environment while the audit is running. The skill, prepared applications, and saved reports provide continuity between runs. A fresh compute node handles the next job.
Temporary compute for a practical task
The demonstration ran within the account's free noBGP allowance, including compute and storage. That describes this experiment's usage; the amount of work another account can run depends on its available allowance and workload.
The broader benefit is the lifecycle. The assistant can request a machine when a task needs one, use it, preserve the useful output, and terminate it when finished.
An SEO audit makes that pattern easy to see. The same approach could support a test run, a data conversion, or an analysis job with a defined finish point. Each needs a place to execute and a way to keep its results.
Give your AI a job it can finish
This project began with a tool evaluation and ended with a reusable audit workflow. Claude deployed the candidates, helped compare their output, and packaged the chosen approach into a skill that could run again.
noBGP supplied the remote execution tools, temporary compute, and storage that made the workflow possible.
Start with a task you already repeat. Define the input, the work to run, and the output you need to keep. Then give your assistant a place to execute it.
Watch the demonstration or set up noBGP with your AI assistant.
