Lastest — Visual Regression Testing

This skill teaches the agent to drive a Lastest instance — running visual regression tests, inspecting screenshot diffs, and approving or rejecting baselines — through the official @lastest/mcp-server MCP server.

When to use

Use this skill when the user:

• Asks to run visual regression tests, UI snapshot tests, or screenshot comparisons
• Mentions "Lastest", "baseline screenshots", "visual QA", or "pixel diff"
• Wants to review, approve, or reject visual diffs on a build
• Asks to heal a failing visual test with AI

Prerequisites

The agent needs the @lastest/mcp-server MCP server configured in the user's MCP client. If it is not present, guide the user through setup:

One-liner (Claude Code, stdio):

claude mcp add lastest -- npx -y @lastest/mcp-server@latest \
  --url <LASTEST_URL> \
  --api-key <LASTEST_API_KEY>

Claude Desktop / Cursor (stdio):

{
  "mcpServers": {
    "lastest": {
      "command": "npx",
      "args": [
        "-y",
        "@lastest/mcp-server@latest",
        "--url",
        "https://app.lastest.cloud",
        "--api-key",
        "YOUR_API_KEY"
      ]
    }
  }
}

Remote HTTP (no local Node required):

claude mcp add --transport http lastest https://app.lastest.cloud/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

The API key is created at Settings → Runners & API Access → Create API Key in the Lastest web UI. Tell the user it is shown only once.

Recommended flow

Always verify connectivity before mutating state.

1. Check health — call lastest_status with action: "health". If it fails, ask the user to verify the URL and API key, then stop.
2. Find the repo — call lastest_repo with action: "list". If there are multiple, ask the user which one.
3. Pick tests:

• All tests in an area: call lastest_area with action: "list_tests" and skip to step 4.
• A subset: call lastest_test with action: "list" (or lastest_area action: "list_tests") and confirm which ones with the user.

1. Run — call lastest_run_tests with the repo id and optional testIds. Poll lastest_build (action: "get") until the build finishes, surfacing progress to the user.
2. Review diffs:

• Call lastest_build (action: "get") for the finished build. If there are visual diffs, inspect them with lastest_get_diffs (scope: "build", or scope: "single" per diff).
• For each diff, describe it to the user and ask whether to approve or reject via lastest_decide_diff. Do not approve/reject without explicit confirmation.

1. AI-assisted test creation or healing — if the user wants a new test, call lastest_create_test with a URL and prompt. If an existing test is broken, call lastest_heal_test after confirming.

Safety rules

• Never call lastest_decide_diff with a buildId (approve-all), lastest_test action: "delete", or lastest_decide_diff action: "reject" on a batch without the user explicitly confirming the scope first — these operations are destructive or change shared baselines for the whole team.
• Never create API keys or change team settings from within this skill.
• If lastest_status (action: "health") or any tool returns a 401, stop and ask the user to regenerate their API key; do not retry with a different key.

Tool reference (via @lastest/mcp-server)

Pure CRUD lives on resource tools that take an action (or scope) discriminator; workflow verbs are standalone.

Full docs: Lastest MCP Server wiki.

Troubleshooting