MCP server that bridges Claude Code to the Claude Code Chrome Bridge extension over a local WebSocket (no claude.ai login required).
23 tools with consistent naming (verb_noun pattern: get_, list_, click, fill, etc.) and basic descriptions. All tools have input schemas with type definitions. However, descriptions are terse (avg ~40 chars, well below the 194-char baseline), lack context on WHEN to use each tool, and omit error recovery guidance. Parameter descriptions are minimal. No output schemas documented. Missing idempotency hints for write operations and no confirmation patterns for destructive tools (javascript_eval, upload_file). Tool composition is reasonable but lacks chaining metadata (e.g., list_tabs returns tab IDs but unclear if switch_tab accepts them).
Attach a tab to the current session
Get the status of the Chrome bridge connection
Click on an element on the page
Close a tab in the browser
Fill a form field with text
Fill multiple form fields
Find text on the current page
Get the text content of the current page
Descriptions are terse (avg ~40 chars vs 194-char baseline). Most lack context on WHEN to use the tool, prerequisites, or what the LLM should expect. E.g., 'Get the status of the Chrome bridge connection' does not explain when to call it or what status values mean.
No output schemas documented. LLMs cannot infer what fields list_tabs returns, what structure take_screenshot produces, or what read_console_messages yields. This forces agents to guess and risks malformed downstream calls.
Destructive/high-risk tools (javascript_eval, upload_file, close_tab) lack confirmation patterns or dry-run support. An agent could execute arbitrary JS or delete tabs without safeguards.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Execute JavaScript code in the current page
List all open tabs in the browser
Navigate to a URL in the current tab
Open a new tab in the browser
Press a keyboard key
Read console messages from the current page
Read network requests from the current page
Read the content of the current page
Resize the browser window
Scroll the page
Switch to a specific tab
Take a screenshot of the current page
Type text into the focused element
Upload a file to the current page
Wait for an element to appear on the page
No error handling guidance. Tools do not document what errors are retryable, what user-fixable errors look like, or how to recover. E.g., 'wait_for' timeout behavior is undocumented.
Tool composition unclear. list_tabs returns tab IDs, but it is not documented whether switch_tab, close_tab, and attach_tab accept those IDs or require a different format. Broken chains force discovery detours.