Shiplight AI is an "agent-native" end-to-end testing platform. Its core idea is to hand over the writing and maintenance of tests to your coding agent, letting the agent operate your application in a real browser and generate readable, editable tests.
-
Main purpose Shiplight is used to build and maintain end-to-end test suites. It lets coding agents walk through the application in a real browser like a user, turning the verification process into end-to-end tests and committing them to the code repository in the form of plain-language YAML. The company says its goal is to achieve full coverage within days and near-zero ongoing maintenance.
-
Main capabilities
-
Coding agent integration: Shiplight is installed into your coding agent as an MCP server plus a set of Skills, giving the agent "eyes and hands" in the browser so it can verify UI changes during development and turn those verifications into end-to-end tests.
-
Readable tests: Tests are written in plain YAML in the repository, describing the user's actions. They can be reviewed in a PR or manually fine-tuned in a local debugger.
-
Self-healing tests: When cached locators fail because a button moves or a class name is renamed, Shiplight re-reads the page, parses the original intent, and updates the cache; "AI Fixer" can fix the remaining issues.
-
Run anywhere: A single command can run locally, on Shiplight-hosted CI runners, or on your own infrastructure. The same YAML file runs in all three environments. Results land on pull requests.
-
Create tests from many sources: Your agent can write plain-English YAML tests by walking through the application, from session recordings, tickets, specifications, or test plans in spreadsheets. Example command: /shiplight create-yaml-tests.
-
Failures return fixes, not tickets: Results land on pull requests with logs, screenshots, and trace information. You can have the agent run /shiplight fix, which reproduces the failure, locates the cause, and updates the test; if the problem is in the application itself, it reports the bug.
-
Build-time verification and review: The same set of browser skills can also be used outside the test suite. Agents can verify changes as they happen and run structured reviews for accessibility, security, and performance, with evidence captured before code review.
-
Gets smarter with every run: Each run distills observed application behavior into reliable knowledge (flows, terminology, what is stable, what changes).Stable steps are deterministically replayed from cache with no model calls; only newly appearing or changed parts require AI.
-
How to use it Installing Shiplight into a coding agent takes just one line. Local browser automation and test writing do not require a Shiplight account or API token, so in-progress changes can be verified before signing up. After registering an account, you get hosted CI runners, result history, and a dashboard. Shiplight is based on Playwright: each YAML test is transpiled at runtime into a standard Playwright test, so it supports all Playwright browsers and capabilities, and cached steps replay at full Playwright speed. It can coexist with existing Playwright suites, with no lock-in: you can exit at any time, and the transpiled tests are standard Playwright files that continue to run even without Shiplight. The official documentation explicitly states that Selenium is not involved.
-
Suitable scenarios It is suitable for engineering teams that need end-to-end test coverage and want to reduce the cost of writing and maintaining tests. The official site shows scenarios in which multiple teams, after using it, gained coverage of critical flows more quickly, reduced manual checks, and caught issues before regressions reached staging. It is also suitable for enterprise teams; the official site says the platform supports SOC 2 compliance, audit logs and access control, private cloud and VPC deployment, a dedicated CSM, and priority support.
-
Limitations The materials do not provide specific pricing, the specific scope of the free plan, the complete list of supported coding agents, or compatibility details for each language or framework, so these are not available for now. The official site mentions that local browser automation and test writing do not require an account or API token, and that the account is used for hosted CI, result history, and the dashboard; you can bring your own LLM key and run it on your own infrastructure (any plan, including the free plan).
