When other browser automation MCP tools such as Playwright or Chrome Devtools are installed, they usually work as expected. However, whenever the word ‘browser’ is mentioned in the prompt, it automatically enables the new built-in Browser Automation mode, which also automatically disables the other tools sometimes (usually Playwright). For our workflows we require running the browsers in headless mode on a Remote-SSH-connected machine with a specific SSL certificate setup, so Browser Automation doesn’t work at all. Thus, this behavior is just disruptive with no benefit.
Please fix this by adding a global disable for Browser Automation so we can at least manually avoid the disruption and blockage of the MCP tools that work for our use case. Ideally, also make your integration smarter by providing Browser Automation as a normal tool just like the other MCP tools, not one that will automatically step in and override others in easily-triggered scenarios like mentioning the common word ‘browser’.
Steps to Reproduce
Connect to a remote development environment with Remote SSH (probably not required).
Install the Playwright and Chrome Devtools MCP servers and ensure they’re working correctly.
Ask an agent to visit a website and take a screenshot using Playwright. Confirm that it works.
Ask an agent to visit a website and take a screenshot using Playwright, but mention the word ‘browser’ anywhere in the prompt as well.
Expected Behavior
Agent successfully uses Playwright in both cases.
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
In the Cursor chat pane, simply typing the word “browser” automatically triggers Cursor’s internal browser tools. The globe icon lights up and the selected tool switches to Browser or Google Chrome, even when I do not intend to use them. This behavior is based solely on the keyword “browser,” and it occurs every single time.
This creates a conflict because I rely on the Playwright MCP tool for all browser automation. My MCP is configured as follows:
When Cursor auto-selects its internal browser tools, the LLM becomes confused because it sees both the internal tools and the Playwright MCP tool available. As a result, it often sends commands to the wrong tool instead of Playwright MCP. There is currently no setting to disable this automatic tool switching, so I’m forced to manually deselect the internal browser tool each time, which interrupts and slows down my workflow.
Steps to Reproduce
Open the chat pane in Cursor.
Type the word “browser” in any message.
Cursor automatically highlights the globe icon.
The internal Browser tool becomes selected.
Expected Behavior
Typing the word “browser” should not cause Cursor to auto-select any internal browser tools. Cursor should not switch tools based on keywords, especially when the user has their own Playwright MCP tool installed.
For AI issues: add Request ID with privacy disabled
Additional Information
This is extremely annoying because I use the Playwright MCP tool all day. Typing the word “browser” is part of my normal commands, and every time I do, Cursor selects an internal browser tool. I have to keep manually unselecting it, and there is no setting to disable this behavior.
Hey @Jon_Rajavuori@shaggy2626 thanks for the report. I can confirm this is a real issue affecting users with custom MCP browser automation tools. I’ll pass this to the team.
