Portelon

Sign in Account

Connecting agents. Portelon connects ten agents in one click. Any other agent that supports MCP takes one command.

In one click

The Agents page lists the agents on your computer, each with a switch. Open it from Agents in Portelon’s menu.

Agent Id
Claude Code claude-code
Codex, and the ChatGPT desktop app codex
OpenCode opencode
GitHub Copilot github-copilot
Cursor cursor
Antigravity antigravity
Devin, formerly Windsurf devin
Cline cline
Zed zed
Kimi Code kimi-code

Connecting an agent adds Portelon to it, and turns off the agent’s own web search where its settings allow, so that it searches through Portelon, with your search engine. Changed files, under the agent, lists what changed. An agent that’s already running picks up Portelon when it next starts.

When you install another of these agents, Portelon offers to connect it the next time you open Portelon.

Disconnecting

Turn the agent’s switch off, or choose Disconnect all. Portelon puts back everything it changed, and leaves any setting you’ve changed since as you left it. Uninstalling Portelon disconnects every agent too.

Any other agent

Any agent that supports MCP can use Portelon. On the Agents page, Other agents shows the command to add to it as an MCP server, with a Copy button. It runs over stdio, needs no arguments, and stays in place through updates.

  • If the agent limits how long a tool call may run, allow ten minutes.
  • In a JSON file on Windows, double each backslash in the command.
  • In the agents above, the entry named portelon is Portelon’s own, so name yours something else.
  • Uninstalling Portelon leaves the entries you added. Remove them yourself.

Options

Add these to the command’s arguments. Without them, agents use your own profile.

  • --profile NAME gives the agent a profile of its own, with its own logins, kept between runs.
  • --ephemeral-profile gives it a fresh profile, deleted when the browser closes.
  • --unattended is for agents no one watches. They can’t show you a tab, give you one, or use your tabs.
  • --fs-root DIR keeps Portelon’s file reads, writes and uploads, and file:// pages, inside a folder that exists. Everything using the profile must give the same folder, so give the agent a profile of its own.
  • --require-sandbox refuses to start unless every layer of the sandbox that agents’ code runs in is in force.
  • --password-store basic works on Linux without a keyring, but saved passwords are effectively unencrypted.

Each also works as an environment variable, named after it: --profile work as PORTELON_PROFILE=work, --unattended as PORTELON_UNATTENDED=1.

From a terminal

  • portelon-mcp agents lists your agents.
  • portelon-mcp agents connect codex claude-code connects the ones you name, by their ids above.
  • portelon-mcp agents disconnect --all disconnects them all.
  • portelon-mcp agents --help lists the rest.

On Linux, portelon-mcp is on your PATH. Elsewhere, run the command the Agents page shows in its place.

When an agent can’t use Portelon

  • Restart the agent. Agents read their settings only when they start.
  • A license. Without one, the agent tells you what to do, such as approve a sign-in code.
  • portelon-mcp doctor checks Portelon’s setup on this computer.