[FEATURE] MCP integration between Claude Code (terminal) and Claude browser extension
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
Currently, Claude Code (terminal) and the Claude browser extension are completely separate tools with no way to communicate with each other. Developers who use both tools in parallel have to manually switch context between the terminal and the browser, copy-paste instructions, and coordinate actions themselves.
A common workflow looks like this:
- Ask Claude Code in the terminal to implement a feature or fix a bug
- 2. Manually switch to the browser to test the result
- 3. Manually describe what happened back to Claude Code
This manual handoff is inefficient and breaks the agentic loop. Claude Code cannot instruct the browser extension to perform actions, and the browser extension cannot report results back to Claude Code. There is no shared context or communication channel between the two.
Proposed Solution
Expose the Claude browser extension as an MCP (Model Context Protocol) server so that Claude Code in the terminal can connect to it and delegate browser actions.
This would enable a fully agentic loop, for example:
User → Claude Code (terminal): "Build this feature and test it in the browser"
Claude Code → writes code, runs dev server
Claude Code → calls browser extension via MCP: "Navigate to localhost:3000, click X, verify Y"
Browser extension → executes actions, returns result
Claude Code → receives result, iterates or reports back to user
Key capabilities this would unlock:
- Claude Code could instruct the browser extension to navigate to URLs, click elements, fill forms, and take screenshots
- - The browser extension could return results (screenshots, page content, errors) back to Claude Code
- - - A single user request in the terminal could trigger a full build + browser test cycle autonomously
- - - - Developers working on local web apps (e.g. localhost) would benefit enormously from this closed feedback loop
Suggested implementation approach:
- The browser extension exposes a local MCP server (e.g. on a localhost port or via a named pipe)
- - Claude Code connects to it as an MCP tool, just like any other MCP server
- - - The MCP tool exposes actions like
browser_navigate,browser_click,browser_screenshot,browser_get_content
This aligns perfectly with Anthropic's existing MCP standard and would be a natural bridge between the two flagship Claude developer tools.
Alternative Solutions
_No response_
Priority
Critical - Blocking my work
Feature Category
CLI commands and flags
Use Case Example
_No response_
Additional Context
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗