chrome-devtools

Navigate to the page with navigate_page or new_page, wait for content with wait_for, then call take_snapshot to get element uids. Use those uids with click, fill, and other input tools to interact with the page.

Follow the sequence: navigate, wait, snapshot, then interact. You can send multiple tool calls in parallel, but maintaining this order ensures elements exist and the page state is current before interaction.

Use take_snapshot for automation since it returns a text-based accessibility tree with element uids, which is faster. Use take_screenshot only when visual inspection of the rendered page is needed.

The element may have been removed or the page changed since the last snapshot. Take a fresh snapshot with take_snapshot to get updated element uids before retrying the interaction.

No, this skill explicitly does not apply to --slim mode MCP configurations. It requires the full chrome-devtools-mcp server, which launches Chrome automatically on the first tool call using a persistent profile.

Use the filePath parameter to write large outputs like screenshots, snapshots, and traces to disk. Apply pagination with pageIdx and pageSize, filter by types, and set includeSnapshot to false on input actions.