Playwright MCP
Let the agent open a browser and check its own work.
Gives an agent real browser control — navigate, click, type, read the accessibility tree, capture console errors. This is the single highest-leverage MCP server for frontend work, because it turns 'it should work' into the agent actually looking.
Best for
Closing the loop so the AI verifies the UI instead of assuming.
Watch out for
Browser automation is slow. Use it to verify, not to explore aimlessly.
Using this and stuck anyway?
Knowing the tool isn't the same as finishing the project. If yours has stopped moving, we'll take a look at it.
Tell us what's broken