Opik

Opik is the open-source LLM observability and evaluation platform from Comet. This connector lets Claude work with your Opik Cloud workspace: find what broke in your production agents, see why, and record what you decide.

What Claude can read

• Project health: traces, error rate, latency and cost for this week against the week before, plus the score names to filter on.
• Diagnostics issues: recurring failures that Opik has already grouped and ranked, with the cause, a suggested fix and example traces.
• Traces, spans and threads, with filters, sort, time windows and free-text search. A trace comes back with its spans, and a thread with its messages.
• Experiments, datasets and test suites, prompts and their versions, online evaluation rules, annotation queues and metrics over time.

What Claude can change, when you ask

• Add feedback scores and comments to traces, spans and threads.
• Create datasets, test suites, dataset items, experiments and prompt versions.
• Close or reopen threads, resolve or close Diagnostics issues, and turn on Diagnostics for a project.
• Create and update annotation queues.

Every answer links to the record in the Opik UI and states its size. When a list or a trace is long, the answer says how much it left out and gives the call that gets the rest.

The connector also serves Opik's agent skills, which show Claude how to instrument an app, run evaluations, compare experiments and turn a failing trace into a regression test.

You sign in with your Comet account (OAuth). There is no API key to set, and Claude works in the workspace that you sign in to.