TestivAI Open Source

What TestivAI Open Source does

The TestivAI Open Source MCP server gives AI coding agents access to the outputs of TestivAI's local visual regression workflow. Its focus is not browser automation by itself, but making captured visual-test evidence available for agent inspection. The available material identifies verdicts, diff images, and the explain_snapshot capability as the core MCP-facing functions.

TestivAI compares screenshots with page DOM snapshots and computed styles. That evidence lets the generated report distinguish several cases: a genuine DOM change, a style-only change, and a pixel difference that is likely caused by rendering noise when the DOM and styles remain identical. Reports can also describe changed regions, selector-level changes, and layout shifts.

How it works

A test or capture run creates local artifacts in a visual-report directory. The run includes results.json, which contains per-snapshot diff percentages and DOM change summaries, along with an HTML report and image data used to inspect differences. The MCP layer makes the relevant results available to an agent instead of requiring the agent to inspect screenshots manually.

explain_snapshot supplies layered evidence for a selected snapshot. That evidence can include selectors that shifted or changed, style-only differences, and detection of a whole-page movement. The language model supplies the final narrative; the project does not require a hosted AI service to produce the explanation.

The underlying workflow supports Playwright, WebdriverIO, Selenium, and other adapters listed by the project. It can also capture pages without a test framework through the testivai witness command. The MCP server is most useful after one of these workflows has produced local visual results for an agent to review.

Setup and configuration

The project states that the MCP workflow needs no account, API key, or network access. This makes it suitable for local development and sandboxed agent environments. Baselines and reports remain on the machine, and baselines can be stored in the project repository.

The provided material does not include the MCP package's exact installation command or a client configuration block. It does identify the MCP implementation as the @testivai/mcp package in the repository. Visual capture itself can be configured with options such as the report directory, pixel threshold, maximum tolerated diff percentage, animation stabilization, ignored selectors, and automatic report opening.

Tools and capabilities

The TestivAI Open Source MCP server provides access to:

• Visual regression verdicts for captured snapshots.
• Diff images for reviewing changed pixels and regions.
• explain_snapshot evidence for agent-generated explanations.
• Machine-readable results from visual-report/results.json.
• DOM change summaries, including added and removed elements.
• Style-change, selector-attribution, mask, and layout-shift evidence when those results were produced by the underlying visual test.

Limitations and notes

This server depends on TestivAI visual results; it is not described as a general browser-control MCP server. The excerpt does not document the full MCP tool schema, transport, command-line arguments, or client-specific configuration, so those details should be verified in the repository before deployment.

Visual baselines still require human judgment. The project’s agent guidance says an agent should report intended changes and should not approve a baseline itself. A human can approve baselines through the project’s GitHub workflow using a /testivai approve comment.