Does chrome-devtools-mcp have a CLI for agent-authored browser scripts?
I'm looking for a way for coding agents to batch browser actions instead of making one MCP tool call per step. The chrome-devtools-mcp repo had an RFC proposing a
chrome-devtools-cliwrapper with bash loops, waits, and screenshots. Was it implemented, and can agents run multiple browser instances in parallel?Originally asked on GitHub: https://github.com/ChromeDevTools/chrome-devtools-mcp/issues/797
5 answers
Implemented. The RFC was closed as completed, and a CLI is now available: a contributor pointed to the skill documentation at https://github.com/ChromeDevTools/chrome-devtools-mcp/blob/main/skills/chrome-devtools-cli/SKILL.md (opens external site). The idea was to let agents write bash scripts that call CLI commands such as
navigate_page,wait_for, andtake_screenshot, instead of making one MCP tool call per browser action.For parallel execution, a contributor stated that running multiple instances in parallel should already work with the
--isolatedflag, though no concrete test output is shown in the thread. The discussion also referenced https://github.com/thomasdondorf/puppeteer-cluster (opens external site) for parallel browser work and https://github.com/vercel-labs/agent-browser (opens external site) as a related project.If you are evaluating this today, the most reliable path is to read the SKILL.md above and try the commands in a terminal outside of an agent first; that is general advice, not something proven in this thread. Another option mentioned in the thread is https://github.com/microsoft/playwright-cli (opens external site), which the commenter positioned as a token-efficient CLI alternative to MCP for coding agents.
No security issues were raised in this thread. The
--isolatedflag name comes from the contributor comment and may also appear in the skill docs, so verify the exact flag spelling in SKILL.md before relying on it.Did this work for you?What @natorion tried on GitHub (Outcome unclear)
A contributor said parallel browser instances should already work with the
--isolatedflag, but the thread does not show a confirmed run.Did this work for you?What @nucliweb tried on GitHub (Outcome unclear)
A commenter pointed to https://github.com/microsoft/playwright-cli (opens external site) as a CLI-based, token-efficient alternative for coding agents, but did not report using it in this thread.
Did this work for you?What a GitHub user tried on GitHub (Outcome unclear)
The issue author proposed a
chrome-devtools-cliwrapper and showed a bash script looping over URLs withnavigate_page,wait_for, andtake_screenshot, but this was a feature proposal, not a tested attempt.Source: https://github.com/ChromeDevTools/chrome-devtools-mcp/issues/797 (opens external site)
Did this work for you?What @BogdanCerovac tried on GitHub (Outcome unclear)
A commenter suggested using puppeteer-cluster for parallel browser tasks and asked whether the CLI could abstract parallel runs, without confirming results.
Did this work for you?
Related questions
- Why does chrome-devtools-mcp snapshot miss table/ARIA semantics?
- Why is --load-extension ignored by chrome-devtools-mcp?
- How do I use chrome-devtools-mcp from WSL when Chrome is installed on Windows?
- chrome-devtools-mcp --autoConnect times out on every tool call with Chrome 144
- How to select which Chrome profile chrome-devtools-mcp automates when several are open?