Hello nf-core community, I built nf-tui, an open source terminal UI for debugging past and currently running Nextflow pipelines. nf-tui reads what Nextflow already writes: .nextflow.log and the run's work directories, straight from disk. No plugin, no agent, no account, no instrumentation. Works on the run going right now, the one a colleague launched, or one that failed in CI last week. Nothing leaves your infrastructure. No run metadata, no sample names, no file paths sent to a SaaS. It's a local process reading local files. Under GxP, PHI, or IP constraints, that removes the issues of sharing data with a third-party service. Nothing to deploy either. No service, no database, no seats to buy. A Nextflow user installs this in their home directory or on a HPC cluster. Debugging is all on one screen: task tree, the selected task's log, output files with sizes as they're written, the run log, and a running-vs-queued view and ability to view files of a task. BAM/CRAM get decoded with samtools from that task's own container. I also built this to be an MCP, so it talks to your agents. nf-tui-mcp serves the run over MCP, so Claude or any MCP client can call get_failures and get every broken task with its root cause, Nextflow's error report, and its .command.err in one call. Triage without a human opening a dashboard. site: stavgrossfeld.github.io/nf-tui Happy to hear feedback and whether this is useful for the community. Best, Stav

Excited about it, I'll have a look. Thanks Stav

Thanks Maxime, glad you came across it! Would love to hear your thoughts once you've had a chance to try it out

Great tool. I’ll try it on my HPC. Python tools tend to be less responsive on lightly resourced jump nodes on our HPC. Compiled tools tend to perform better.

Lately with the raise of coding TUIs I've just realized how important and useful those are. This is very similar to the one I built for the horus-runtime. The MCP piece I don't understand though. Why not just edit the workflow or let the LLM read the log files directly to propose fixes? Would be great to connect and swap notes sometime, so I can understand better your implementation details 💪