From helium-browser
Control and debug the user's local Helium browser (helium.computer) via DevTools MCP. Use when automating Helium, testing web apps in the user's real browser, taking screenshots, inspecting network/console, profiling performance, or debugging pages the user already has open. Triggers on "helium browser", "my browser", "local browser", "helium tab", or when helium MCP tools are available. Do NOT use for headless scraping of arbitrary sites — prefer firecrawl for that.
How this skill is triggered — by the user, by Claude, or both
Slash command
/helium-browser:helium-browserThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Connects to the **running Helium browser on this machine** via `helium-devtools-mcp` (a Helium-patched fork of chrome-devtools-mcp). Tools are exposed as `helium__*` (e.g. `helium__navigate_page`, `helium__take_snapshot`).
Connects to the running Helium browser on this machine via helium-devtools-mcp (a Helium-patched fork of chrome-devtools-mcp). Tools are exposed as helium__* (e.g. helium__navigate_page, helium__take_snapshot).
Unlike launching a fresh Chrome profile, --autoConnect attaches to the user's existing Helium session — their tabs, cookies, extensions, and logins.
Grok connects to a local MCP gateway that keeps one browser CDP session alive. Start it once per login session:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/start-persistent-server.sh
Approve Helium once when the gateway first connects. Later agent turns reuse the same session — no more prompts.
Either mode works:
A. UI remote debugging (current Helium session): open helium://inspect/#remote-debugging, enable it, approve once when the gateway starts.
B. Classic debug port (zero prompts): close Helium, then launch with:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/launch-helium-with-debug.sh
Run ${CLAUDE_PLUGIN_ROOT}/scripts/status.sh to verify readiness.
Browser lifecycle: The MCP server connects to the running Helium instance; it does not start Helium. If connection fails, check remote debugging first.
Page selection: Tools operate on the currently selected page. Use helium__list_pages, then helium__select_page to switch context.
Element interaction: Use helium__take_snapshot to get page structure with element uids. If an element isn't found, take a fresh snapshot.
helium__list_pages — see what's openhelium__select_page — pick the target tabhelium__take_snapshot — understand page structurehelium__click, helium__fill, helium__navigate_page, etc.filePath for large outputs (screenshots, snapshots, traces)pageIdx, pageSize) to minimize dataincludeSnapshot: false on input actions unless you need updated page statehelium__take_snapshot (text-based, faster)helium__take_screenshothelium__evaluate_scriptMaintain order: list/select page → wait → snapshot → interact. You can parallelize independent read-only calls.
| Task | Use |
|---|---|
| Test a page the user has open | Helium MCP |
| Debug with user's real session/cookies | Helium MCP |
| Scrape arbitrary public URLs | Firecrawl |
| Headless CI-style automation | Firecrawl or chrome-devtools-mcp |
Helium MCP exposes the full contents of the user's browser to the agent. Never exfiltrate passwords, tokens, or personal data. Warn the user before navigating to sensitive pages if appropriate.
npx claudepluginhub awfixers-stuff/plugins --plugin helium-browserSets up isolated workspaces using native worktree tools or git worktree fallback. Use before starting feature work to protect the current branch.