Roadmap
This is where Plainwright is heading after 2.0.0. It’s a direction, not a schedule: there are no dates, and priorities move with feedback. If something here matters to you, say so on the issue tracker, or pick it up (see Contributing).
Next: polish what 2.0 shipped
Section titled “Next: polish what 2.0 shipped”Known limits of the current release, most of them small.
App
- Filter the Runs page by effective status, so overridden runs are listed under their corrected result.
Accounts and API
- Personal API tokens for scripts and CI, so automation doesn’t need a user password, with revocation per token.
- A setting to turn off open registration.
- Per-address sign-in limits that see the real client address behind the bundled proxy.
Runs and evidence
- Cost estimates in alerts, and per-row costs reported by eval targets.
- Root-cause and chat tokens in the run’s own totals and cost.
- Lighthouse-grade performance metrics (today’s numbers come from browser timing APIs, and total blocking time is approximated).
Protocols and integrations
- MCP: contract checks for resources and prompts, and servers that need the MCP OAuth flow.
- A2A: streaming (
message/stream), push notifications, and the gRPC and HTTP+JSON transports. - Jira Server and Data Center, and Linear’s OAuth sign-in.
- Protection against DNS rebinding in the outbound-request guard, and a network policy that keeps pages under test away from the backend and workers too.
Self-hosting
- A production profile for Docker Compose that also drops the debugging ports bound to
127.0.0.1. - Deploy guides verified end to end on each platform.
Later: bigger features
Section titled “Later: bigger features”Ideas we want to build, roughly in order of how often they come up.
- Playwright traces and video for browser runs, alongside today’s screenshots and logs.
- Take control of a live run: pause, solve a step by hand (an SMS code, a CAPTCHA), then let the agent continue.
- Data-driven UI tests: run one plain-English test over the rows of a dataset.
- Self-healing reports: when a test passes despite a changed page (“Sign in” is now “Log in”), say so and suggest updated wording.
- Test impact: run only the tests related to what changed in a deploy.
- Anomaly detection: flag unexpected changes in the page or in AI answers without an explicit assertion.
- Requirements traceability: from an imported issue or spec to its tests and their latest results.
- Privacy tools: redaction of personal data in screenshots, data export and account deletion.
- Organisation features: SSO/SAML, audit logs, and workspaces above projects.
- Native mobile apps, beyond the mobile browser device presets.
Recently shipped
Section titled “Recently shipped”See the Changelog for what 2.0.0 delivered: AI-product testing (conversations, red teaming, LLM and RAG evals, MCP and A2A), reliability and triage with automatic issue filing, the CI gate, cost estimates, integrations, security hardening and self-hosting operations. Pages now load on demand, the Crawler’s Stop cancels on the server, the Team page and Settings → Members share one view, and replaced visual baselines are cleaned up in storage.