MCP tools reference

What each grip-mcp tool does, when to use it, and a simple workflow that keeps agents out of trouble.

Overview

These tools are exposed by grip-mcp to any MCP client. They operate on the Chrome tab attached through DevTools Protocol.

Refs from snapshot expire when the page navigates or changes heavily. Always snapshot again before clicking or filling on a new page state.

Not set up yet? Start with the MCP configuration guide.

snapshot()

Read the page accessibility tree and get refs you can use in other tools.

When to use: After every page load or navigation.

highlight()

Draw a blue outline on an element so you can confirm the right target.

When to use: Before click or fill, especially when debugging agents.

click()

Click a button, link, or control by its ref from snapshot.

When to use: When you need to submit, navigate, or toggle UI.

fill()

Type text into an input or textarea.

When to use: Forms, search boxes, login fields.

read_logs()

Read console output the page has written (errors, warnings, logs).

When to use: After an action to see if something broke.

read_network()

List recent network requests as HAR-style entries.

When to use: Debugging API calls or failed fetches.

screenshot()

Capture the full page or a single element as an image.

When to use: Visual verification or sharing state with a user.

pick_element()

Turn on the visual picker — you click an element, Grip returns its metadata.

When to use: When you need human-in-the-loop element selection.

navigate()

Go to a URL. All previous refs become invalid after navigation.

When to use: Moving to a new page in a flow.

eval()

Run JavaScript in the page context via Runtime.evaluate.

When to use: Advanced cases not covered by click/fill.

A reliable pattern for agents (and humans driving MCP manually):