Currents: Currents for Claude & Codex to see all your tests
A test results dashboard for Playwright and Cypress now lets Claude and Codex read its continuous integration data. Users connect from the Claude or ChatGPT and Codex plugin directories, sign in, and ask an agent why a pipeline failed. The agent then receives the failure, the history of that test and the trace, automatically. The launch appeared on Product Hunt on 20 November 2026 with 156 upvotes and 24 comments at the time of reporting.
Currents records each Playwright run in CI, and its reporter sends every result to the dashboard in real time. The connection runs through a hosted MCP server at mcp.currents.dev, which lets agents query runs, tests and analytics on demand. Beyond reading, a connected agent can act on results, for example by quarantining a flaky test, per Currents' own guide. The company also describes a fix workflow in which an agent reads a failed run and trace, reproduces the failure locally, and pushes changes until CI passes.
Two caveats apply. A draft documentation pull request in Currents' readme repository described the ChatGPT and Codex listing as pending OpenAI approval, so buyers should confirm it is live. The company's site also cites customer results, such as a 90% cut in test flakiness at n8n, figures it reports itself and that were not independently verified here.
Why it matters
For companies building developer tools, the agent connection now runs through platforms owned by OpenAI and Anthropic. Plugin directories give a vendor distribution inside ChatGPT, Codex and Claude, but they also make approval and listing rules a dependency. On 9 July 2026, OpenAI replaced its app directory with a plugin directory that also serves Codex, so one listing can reach both products. Investors evaluating dev tools should ask whether a startup owns the data agents consume, such as run history and traces, or only a front end on someone else's model.
For operators, the practical question is where test history lives. An agent is only as useful as the run data behind it, so retention across repositories and CI systems becomes a selling point. Buyers will also ask what leaves their network, because traces can contain page content and session details captured during tests. Founders selling into engineering teams should expect buyers to request agent access as a default, and should plan for the usage costs that follow.
Key points
- 1Users connect from the Claude or ChatGPT and Codex plugin directories, then sign in through their browser.
- 2The agent receives the failure, that test's run history and the Playwright trace for the failed run.
- 3The launch had 156 upvotes and 24 comments on Product Hunt when the story was reported.
- 4The plugin connects to a hosted MCP server at mcp.currents.dev, according to a draft Currents readme change.