About
I connect agents to real browsers. Playwright does the clicking. An MCP server is the door: the agent asks for a tool, the server runs it, and the page either moves or it doesn’t.
Playwright is the part that tells the truth about the page. A locator, a trace, a worker count. The run that only goes red once several browsers start together. An MCP server is how an agent reaches that, and the files and commands around it, through one protocol instead of a fresh script every time. Agentic only counts when the tool result comes back and the page matches it.
The agent will say the editor is filled. The editor still shows the stub. The posts stay with that kind of failure until the mechanism is obvious. Company and product names stay out.
Code and smaller experiments are on GitHub. I am on LinkedIn. Email is manikrishna4849