Summary
A CI failure becomes something a model can actually read.
A Playwright trace is a binary archive that no assistant can open and that no context window can swallow whole. This server turns it into targeted, paginated answers, so the question "why did this test fail" gets a file and a line number rather than a shrug.
What it is
An MCP server that opens Playwright `trace.zip` archives and exposes their contents as structured, paginated results, with ARIA snapshots compressed so a failing run fits in a model's context.
What you get
- A summary of what the trace contains, so the investigation starts from the shape of the run rather than raw JSON
- The test's own metadata — which spec, which run, which configuration produced this archive
- The failing action resolved back to source code — the spec file, and the line and column, behind a `locator.click` timeout
- Screenshot frames from around the moment of failure, rather than the whole video
- Pagination and compression on every read, so pulling detail does not blow the context window
Requirements
The npm package `playwright-trace-decoder-mcp` and a path to a trace.zip on disk — the same archive CI already publishes as an artifact. No account and no key.
Setup effort
One command — npx -y playwright-trace-decoder-mcp