MCP
Let an agent on your PC use your browser accounts through Computron.
computron mcp is an MCP server on stdio. An agent on the same PC hands it a task in words, such as “list the DNS records for example.com in my Cloudflare account”. Computron does it in your own Chrome, signed in as you, and the agent gets back what happened: the value read, the page it ended on, or what you have to do.
Tools
browser_task: any request in plain words.- One tool per learned function, with the dot replaced:
x.postbecomesx_post. Each is marked read-only or destructive, so an agent can ask before it runs one that changes something.
Each call is the same run as typing the task in the terminal, so it uses your Computron credits the same way. The first call may open your browser to sign in to Computron and agree to its Privacy Policy.
Register it
Register it once. On Windows, if computron came from npm link (a .cmd file), start it through cmd /c computron mcp. The installed computron.exe needs no wrapper.
| Client | How |
|---|---|
| Claude Code | claude mcp add --scope user computron -- computron mcp |
| Codex CLI | In ~/.codex/config.toml, as below. Keep tool_timeout_sec = 600: Codex stops a tool call after 60 seconds by default, and most browser tasks take longer. |
| Cursor | In ~/.cursor/mcp.json, as below. |
| Claude Desktop | Settings, Developer, Edit Config (%APPDATA%\Claude\claude_desktop_config.json, or ~/Library/Application Support/Claude/claude_desktop_config.json on macOS). Add the same mcpServers entry as Cursor, then restart the app. |
Codex CLI:
[mcp_servers.computron]
command = "computron"
args = ["mcp"]
tool_timeout_sec = 600Cursor and Claude Desktop:
{
"mcpServers": {
"computron": { "command": "computron", "args": ["mcp"] }
}
}Make it the default
The agent decides when to use the tool. To make it the first choice for work in your accounts, add a line to your CLAUDE.md (Codex: AGENTS.md):
Browser work in my own accounts (consoles, settings, CLI sign-ins): use computron's browser_task.
Debugging a page I am building: Chrome DevTools.Limits
Calls run one at a time. A call the agent cancels still runs to its end, because stopping between a press and its check is the riskier choice. To stop a run, press Cancel on Chrome's “is debugging this browser” bar.