Claude Desktop

\n\nEdit your Claude Desktop config file:\n- **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json`\n- **Windows**: `%APPDATA%\\Claude\\claude_desktop_config.json`\n\n```json\n{\n \"mcpServers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n }\n}\n```\n\nRestart Claude Desktop after saving.\n\n

\n\n

Cursor

\n\n1. Open Cursor Settings (Cmd/Ctrl + ,)\n2. Go to \"Features\" -> \"MCP Servers\"\n3. Click \"Add Server\" and add:\n\n```json\n{\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n}\n```\n\nOr edit `~/.cursor/mcp.json` directly.\n\n

\n\n

VS Code (GitHub Copilot)

\n\nAdd to your VS Code settings.json:\n\n```json\n{\n \"github.copilot.chat.mcpServers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n }\n}\n```\n\n

\n\n

Windsurf

\n\nEdit `~/.codeium/windsurf/mcp_config.json`:\n\n```json\n{\n \"mcpServers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n }\n}\n```\n\n

\n\n

OpenCode

\n\nAdd to your `.opencode/config.json`:\n\n```json\n{\n \"mcp\": {\n \"servers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n }\n }\n}\n```\n\n

\n\n

Gemini CLI

\n\nAdd to `~/.gemini/settings.json`:\n\n```json\n{\n \"mcpServers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\"]\n }\n }\n}\n```\n\n

\n\n

OpenAI Codex CLI / ChatGPT desktop app / Codex IDE extension

\n\nAll three share one config file: `~/.codex/config.toml`. Configure once, use everywhere.\n\nEasiest — let Codex write the entry:\n\n```bash\ncodex mcp add vibe -- npx -y @vibebrowser/mcp\n```\n\nOr add the table to `~/.codex/config.toml` by hand (Codex uses **TOML**, not JSON):\n\n```toml\n[mcp_servers.vibe]\ncommand = \"npx\"\nargs = [\"-y\", \"@vibebrowser/mcp\"]\n```\n\nIn the ChatGPT desktop app you can also use the UI: **Settings → MCP servers → Add server**, choose **STDIO**, command `npx`, args `-y @vibebrowser/mcp`, then **Restart**.\n\nVerify with `codex mcp list`, or type `/mcp` in the Codex TUI or the desktop composer.\n\n> ChatGPT on the web does not read local Codex config, so a local MCP server like Vibe is not available there.\n\n

\n\n### MCP command forms\n\nUse the shortest direct package invocation for the MCP server:\n\n```bash\nnpx -y @vibebrowser/mcp@latest --help\nnpx -y @vibebrowser/mcp@latest start --transport http\nnpx -y @vibebrowser/mcp@latest openclaw --remote \"$VIBE_REMOTE_URL\"\n```\n\nBackward-compatible aliases still work when you need explicit binaries:\n\n```bash\nnpx -y -p @vibebrowser/mcp@latest vibebrowser-mcp --help\nnpx -y -p @vibebrowser/mcp@latest vibe-mcp --help\n```\n\n### Remote connector (hosted assistants — no install)\n\nEverything above assumes the client can spawn a local process. Hosted\nassistants cannot: Claude on the web / in Cowork / on mobile, and ChatGPT on\nthe web, run in the vendor's cloud with no access to your machine. They accept\na **remote MCP server URL** and nothing else — no command, no arguments, and\nno custom request headers.\n\nFor those, skip `@vibebrowser/mcp` entirely. The extension alone is enough:\n\n```\nhttps://relay.api.vibebrowser.app/mcp/\n```\n\nFind `` in the extension: **Vibe icon → Settings → AI Agent\nControl → Remote (internet) → Relay access**. That panel shows this exact\nconnector URL, and the same string is now the **preferred** value for the\nCLI/server `--remote` flag — paste it in both places, no translation needed:\n\n```bash\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nnpx -y @vibebrowser/mcp@latest start --remote \"$VIBE_REMOTE_URL\"\n```\n\nA bare extension UUID or a `wss://relay.api.vibebrowser.app/` relay URL\nare still accepted as advanced/compatibility forms — see the accepted\n`--remote` values table in \"Cloud OpenClaw -> Local Browser\" below.\n\nThis is a plain **Streamable HTTP** MCP endpoint. There is **no OAuth consent\nflow, no dynamic client registration, and no scope setup** on this path — the\nUUID in the URL (or header, below) is the entire credential. If a client insists\non an OAuth login before it will add the server, that client is not supported\nhere; use the local stdio path instead.\n\nWhere to paste it:\n\n\n| Client | Where |\n|---|---|\n| Claude (web, Cowork, mobile) | Settings → Connectors → Add custom connector |\n| Claude Desktop — **alternative only**; prefer the local stdio entry above when Chrome runs on the same machine | Settings → Connectors → Add custom connector |\n| ChatGPT (web) | Settings → Connectors → developer mode |\n| Codex Desktop / Codex CLI (remote form) | `codex mcp add vibe --url https://relay.api.vibebrowser.app/mcp/` |\n\n\nCustom connectors are a paid-plan feature in the Claude and ChatGPT products.\n\n> **Codex Desktop, Codex CLI and the Codex IDE extension can spawn a local\n> process**, and they all share `~/.codex/config.toml`. Prefer the local stdio\n> entry above — it keeps traffic on `127.0.0.1` and keeps the UUID out of a URL.\n> Use `codex mcp add vibe --url …` only when the browser you want to drive is on\n> a different machine from Codex. The ChatGPT **desktop app** is Codex-backed and\n> reads the same config, so it is a local-stdio surface too; only ChatGPT **web**\n> needs the hosted URL.\n\n#### Migrating away from the retired OAuth-style connector guidance\n\nEarlier drafts of these docs (and the Aug 2026 directory-submission research\nkept under `worklog/`) described a retired OAuth 2.1 + DCR consent flow at a\nbare `/mcp`. That path is **retired and unsupported for users**. If you\npreviously added Vibe as a connector and were sent to a retired consent screen,\nremove that connector and re-add it with the direct\n`https://relay.api.vibebrowser.app/mcp/` URL above. Nothing to authorize,\nnothing to register, no scopes to pick.\n\nThe relay also accepts the UUID as an `X-Remote-Session` or\n`Authorization: Bearer` header on a bare `POST /mcp`. Use the header form from\nanything that can send one (Codex CLI, scripts) — it keeps the credential out\nof URLs, logs, and browser history. The path form exists specifically for the\nUIs that cannot send a header.\n\n**Two things this path costs you, stated plainly:**\n\n1. **The URL is a credential.** Anyone who has that UUID can drive your\n logged-in browser — read your mail, act as you on any site you are signed\n into. Treat it exactly like a password: never commit it, never paste it into\n a shared chat, an issue, a README, or a screenshot. If it leaks, revoke the\n session in the extension and generate a new one. (This is not hypothetical:\n a dev machine's UUID once shipped in these very docs.)\n2. **Page content leaves your machine.** On the local STDIO path, tool calls\n and page content never leave `127.0.0.1`. On the remote-connector path they\n traverse `relay.api.vibebrowser.app`, because the assistant is in the\n vendor's cloud and has no other route to your browser. If you need\n on-device-only, use a desktop client with the local server instead.\n\n### 3. Connect the Extension\n\n1. Open Chrome with the Vibe extension installed\n2. Click the Vibe extension icon in the toolbar\n3. Go to Settings and enable \"MCP External Control\"\n4. The status should show \"Connected\"\n\nIf the extension is not connected, `vibebrowser-mcp` can optionally fall back to\n`chrome-devtools-mcp` (started in `--autoConnect` mode) when that package is installed.\nThis fallback runs once in the shared local relay daemon (multi-agent safe), so\nboth `vibebrowser-mcp` and `vibebrowser-cli` use the same backend instance.\nWhen extension is connected, extension tools are authoritative.\n\n### `--devtools` mode (drive your real Chrome over CDP)\n\nPass `--devtools` to either CLI to bypass relay/extension routing entirely and\ndrive your **real running Chrome** directly over the Chrome DevTools Protocol.\nThis backend (ported from the `chrome-use` skill) reads Chrome's\n`DevToolsActivePort` file and autoConnects to your live profile — no extension,\nno external MCP server, zero extra dependencies. Requires Chrome 144+; the\npermission dialog fires once per profile.\n\n`--devtools` exposes a focused tool set: `navigate`, `snapshot` (accessibility\ntree with `@eN` refs), `click`, `fill`, `type`, `press_key`, `hover`, `scroll`,\n`screenshot`, `eval`, `get_text`, `get_url`, `get_title`, and tab management\n(`list_tabs`, `new_tab`, `select_tab`, `close_tab`). Use `snapshot` to get\n`@eN` element refs, then pass them as selectors to `click`/`fill`/`type`.\n\nOverride the Chrome profile/channel with `VIBE_CHROME_USER_DATA_DIR` and\n`VIBE_CHROME_CHANNEL` (`stable` | `canary` | `beta` | `dev`).\n\nNot yet covered by `--devtools` v1 (addable later as more CDP domains are wired):\nnetwork request inspection, console logs, performance traces, Lighthouse, memory\nsnapshots, device emulation, dialog handling, file upload, and drag.\n\n## Available Tools\n\n| Tool | Description |\n|------|-------------|\n| `navigate_to_url` | Navigate to any URL |\n| `go_back` / `go_forward` | Browser history navigation |\n| `click` | Click elements on the page |\n| `type` / `fill` | Enter text into inputs |\n| `scroll` | Scroll the page |\n| `take_screenshot` | Capture screenshots |\n| `get_page_content` | Extract page text/HTML |\n| `get_tabs` / `create_new_tab` / `switch_to_tab` / `close_tab` | Tab management |\n| `keyboard_shortcut` | Press keyboard combinations |\n| `web_search` | Search the web |\n\n## How It Works\n\nDefault local mode (no flags):\n\n```\nClaude / Cursor / VS Code (stdio)\n │\n ▼\n [vibebrowser-mcp]\n │ ws://127.0.0.1:19888\n ▼\n Local Relay (auto-spawned)\n │ ws://127.0.0.1:19889\n ▼\n Vibe Extension (Chrome)\n```\n\n1. AI applications connect via MCP over stdio\n2. `vibebrowser-mcp` connects to the local relay on port `19888`\n3. The relay forwards commands to the extension on port `19889`\n4. Results flow back to the agent\n\n### Two legs, two protocols\n\nDon't conflate them — the transport to the MCP server is configurable, the transport to the extension is not.\n\n| Leg | Protocol | Configurable? |\n|---|---|---|\n| MCP client → `vibebrowser-mcp` | stdio **or** streamable HTTP | Yes — `--transport stdio\\|http` |\n| `vibebrowser-mcp` ↔ relay ↔ extension | WebSocket only | No |\n\n```\nMCP client ──stdio──┐\n ├──> [vibebrowser-mcp] ──ws──> [relay] ──ws──> [extension]\nMCP client ──http───┘\n```\n\nStreamable HTTP lets a remote or hosted agent — one that can't spawn a stdio\nsubprocess — talk to the server over a URL:\n\n```bash\nnpx -y @vibebrowser/mcp@latest start --transport http\n# serves POST/GET http://127.0.0.1:8788/mcp\n```\n\nDefaults: `--host 127.0.0.1`, `--http-port 8788`, `--http-path /mcp`. Add\n`--allow-host ` (repeatable) if you front it with a proxy or bind it\nbeyond localhost.\n\n### Multi-Agent Mode\n\nWhen multiple agents connect, Vibe MCP automatically spawns a relay daemon:\n\n- First agent starts the relay (listens on ports 19888 and 19889)\n- Additional agents connect to the relay as clients\n- Relay multiplexes all agent requests to the single extension connection\n- Each agent receives only its own responses\n\n### Cloud OpenClaw -> Local Browser\n\n> ⚠️ **Security:** The `--remote` value below is a live credential — a relay URL/UUID/connector URL grants full control of the target browser session. It is the *sole* bearer credential (there is no second-factor token). Treat it like a password: keep it secret, never commit it or paste it into logs, and if it leaks, regenerate it in the Vibe extension Settings.\n\n#### Accepted `--remote` values\n\n`--remote` (and the MCP `set_remote` tool) accept three forms, all normalized to the same relay connection:\n\n| Form | Example | Maps to | Status |\n|---|---|---|---|\n| Connector URL | `https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000` | relay `wss://relay.api.vibebrowser.app` + UUID | **Recommended** — same string shown in extension Settings and pasted into Claude/ChatGPT connectors |\n| Bare extension UUID | `00000000-0000-0000-0000-000000000000` | default public relay + UUID | Advanced / compatibility |\n| ws(s) relay URL | `wss://relay.api.vibebrowser.app/00000000-0000-0000-0000-000000000000` | explicit relay endpoint | Advanced / compatibility |\n\nSelf-hosted or local relays derive the same way: `https://your-host/vibe/mcp/` → relay `wss://your-host/vibe`; a loopback connector like `http://127.0.0.1:19889/mcp/` → `ws://127.0.0.1:19889` (self-hosted/local relay only — non-loopback hosts must use `https://`/`wss://`).\n\nRejected: an invalid UUID; a URL with embedded credentials, a query string, or a fragment; plaintext `http://`/`ws://` for a non-loopback host; and any HTTP(S) URL that doesn't end in the exact `/mcp/` suffix.\n\nIf your agent runs in the cloud but you want it to control the user's real local browser, run `vibebrowser-mcp` in HTTP mode and connect it to the Vibe extension in remote relay mode. Pass the connector URL (preferred), the extension UUID, or the full WebSocket relay URL to `--remote`.\n\n```bash\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nnpx -y @vibebrowser/mcp@latest start --transport http --remote \"$VIBE_REMOTE_URL\"\n```\n\nThis exposes a local MCP endpoint at `http://127.0.0.1:8788/mcp` by default.\n\nWhen OpenClaw runs on a different machine (for example cloud-hosted), provide a reachable URL:\n\n```bash\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nPUBLIC_MCP_URL=\"https://browser-bridge.example.com/mcp\"\nnpx -y @vibebrowser/mcp@latest openclaw --remote \"$VIBE_REMOTE_URL\" --public-url \"$PUBLIC_MCP_URL\"\n```\n\nYou can print the exact OpenClaw-friendly setup with:\n\n```bash\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nnpx -y @vibebrowser/mcp@latest openclaw --remote \"$VIBE_REMOTE_URL\"\n```\n\nUse the connector URL (preferred) with the default public relay, a bare UUID (default public relay), or a `wss://` URL when you need an explicit relay endpoint. Whichever form you use is the only credential — never share it, log it, or paste it into an untrusted chat.\n\nFor direct browser CLI checks, always use `npx -y @vibebrowser/cli@latest`:\n\n```bash\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" --json status\n```\n\nThe connector URL uses the default public relay. `--remote ` also uses the default public relay. `--remote ` targets an explicit relay endpoint. No second-factor secret is needed or accepted — whichever form you pass is the sole credential that authorizes the session.\n\nFor the full walkthrough, see `docs/openclaw-local-browser.md`.\n\n### OpenClaw-Compatible Browser CLI\n\n`npx -y @vibebrowser/cli@latest` mirrors the OpenClaw browser CLI shape for the real local-browser path:\n\n```bash\nnpx -y @vibebrowser/cli@latest sessions\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" status\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" tabs\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" open https://example.com\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" snapshot\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" click 12\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" type 23 \"hello\" --submit\nnpx -y @vibebrowser/cli@latest --devtools status\n```\n\nUse the MCP package for MCP server workflows and the CLI package for direct browser control:\n\n- `vibebrowser-mcp` for MCP server, HTTP bridge, and helper commands\n- `npx -y @vibebrowser/cli@latest` for OpenClaw-inspired browser control against the real Vibe-connected session\n\n`@vibebrowser/cli` accepts the OpenClaw-style `--browser-profile` flag for compatibility and supports `--json` for machine-readable output. Unlike OpenClaw's managed `openclaw` browser profile, this CLI always targets the real Vibe-connected browser session.\n\nLocal-session selection:\n\n- `npx -y @vibebrowser/cli@latest sessions` lists connected local browser sessions.\n- `npx -y @vibebrowser/cli@latest --session ...` targets a specific local session.\n- If `--session` is omitted in local mode, the CLI uses the first connected session.\n- In remote mode, pass the connector URL (preferred), a bare `--remote ` to use the default public relay, or `--remote ` to use an explicit relay endpoint.\n\nSnapshot behavior is tool-only (no legacy snapshot RPC shortcut):\n\n- `snapshot` (default, `--format ai`) resolves via the `take_md_snapshot` tool — uses the content script's in-page markdown extractor. Fast and readable, but **may return empty for background tabs or complex SPAs** (Notion, Gmail) where the content script is unreachable or layout is not computed.\n- `snapshot --format aria` resolves via the `take_a11y_snapshot` tool — uses Chrome DevTools Protocol `Accessibility.getFullAXTree` directly. **Reliable for all tabs including background tabs and SPAs.** Use this as a fallback when the default format returns empty or only a page title.\n\nThis keeps CLI behavior aligned with extension-supported tools and ensures page targeting works consistently with `--page-id`/`--pageId`.\n\nFor navigation-style operations, responses now include page content when page state changes:\n\n- CLI `open` / `navigate` include `pageContent` in JSON output.\n- MCP tool calls for navigation-style tools return text content that includes current page state (with snapshot fallback when needed).\n\n### Install Prompt for AI Agents (OpenClaw / Hermes)\n\nUse the single canonical copy-paste prompt at the top of this README:\n**[Install prompt for AI agents (OpenClaw / Hermes)](#install-prompt-for-ai-agents-openclaw--hermes)**.\nIt installs the `vibebrowser` skill, saves your remote so you only give it once, and proves\ncontrol of your real Chrome.\n\nFor a deeper skill definition (when to prefer this over a managed/headless browser, the remote-recall rules, selector/snapshot guidance), install `openclaw/vibebrowser/SKILL.md` into the agent's skills directory.\n\n### OpenClaw Integration\n\nThere are two ways to use Vibe with OpenClaw:\n\n**Option A: Cloud OpenClaw controlling local browser**\n\nIf OpenClaw runs in the cloud but you want it to control your local browser:\n\n1. Install the Vibe extension and enable **Remote** mode (see [docs/openclaw-local-browser.md](https://github.com/VibeTechnologies/vibe-mcp/blob/main/docs/openclaw-local-browser.md))\n2. Start the local HTTP bridge: `vibebrowser-mcp openclaw --remote \"$VIBE_REMOTE_URL\" [--public-url \"$PUBLIC_MCP_URL\"]`\n3. Register the MCP URL in OpenClaw\n\n**Option B: OpenClaw skill for local agents**\n\nFor OpenClaw agents that need your real browser context (logged-in sessions, existing tabs):\n\n1. Copy the Vibe skill from this package to your OpenClaw skills folder\n2. Use the connector URL (preferred), the extension UUID, or full WebSocket relay URL with `--remote`\n3. Use `npx -y @vibebrowser/cli@latest` commands in your agent prompts\n\nThe skill is located at [`openclaw/vibebrowser/SKILL.md`](https://github.com/VibeTechnologies/vibe-mcp/blob/main/openclaw/vibebrowser/SKILL.md) and provides:\n- Full OpenClaw-compatible CLI commands (`status`, `tabs`, `snapshot`, `click`, `type`, etc.)\n- Fallback-safe commands for DevTools-backed flows (`resize`, `upload`, `dialog`)\n- `--json` output for machine parsing\n- Environment-based configuration\n\nSee [docs/openclaw-local-browser.md](https://github.com/VibeTechnologies/vibe-mcp/blob/main/docs/openclaw-local-browser.md) for the complete walkthrough.\n\n## Local LLM: `serve` Command\n\nRun a local LLM with one command — no cloud API keys required. Automatically installs [Ollama](https://ollama.com), downloads the model, and starts serving an OpenAI-compatible API.\n\n```bash\nnpx -y @vibebrowser/mcp@latest serve qwen3.5\n```\n\nThat's it. Works on **macOS**, **Linux**, and **Windows**.\n\n### What it does\n\n1. **Detects Ollama** → installs it if missing (via `brew`, `curl`, or `winget`)\n2. **Starts the server** → launches `ollama serve` in the background\n3. **Downloads the model** → streams download progress to your terminal\n4. **Prints connection info** → ready to use with VibeBrowser or any OpenAI-compatible client\n\n### Recommended models\n\n```bash\nnpx -y @vibebrowser/mcp@latest serve qwen3.5 # Best overall for agentic tasks\nnpx -y @vibebrowser/mcp@latest serve llama4 # Strong general reasoning\nnpx -y @vibebrowser/mcp@latest serve deepseek-r1 # Reasoning chains\nnpx -y @vibebrowser/mcp@latest serve mistral # Lightweight & fast (7B)\n```\n\n### Options\n\n```text\nnpx -y @vibebrowser/mcp@latest serve [options]\n\nOptions:\n -p, --port Ollama API port (default: 11434)\n -y, --yes Skip install confirmation prompts\n -d, --debug Enable debug logging\n```\n\n### Using with VibeBrowser extension\n\nAfter `serve` completes, configure the extension:\n- **Model provider** → `ollama`\n- **Model name** → the model you served (e.g., `qwen3.5`)\n\nThe extension connects to `http://localhost:11434/v1` automatically.\n\n## CLI Options\n\n```text\nnpx -y @vibebrowser/mcp@latest --help\nnpx -y @vibebrowser/cli@latest --help\n\n# MCP server (default)\nnpx -y @vibebrowser/mcp@latest [start] [options]\n -p, --port WebSocket port for local relay (agent) connection (default: 19888)\n -d, --debug Enable debug logging\n --transport MCP transport to expose: stdio or http (default: stdio)\n --host Host to bind the HTTP server to (default: 127.0.0.1)\n --http-port Port for streamable HTTP MCP transport (default: 8788)\n --http-path Path for streamable HTTP MCP transport (default: /mcp)\n --allow-host Allowed host header for HTTP transport (repeatable)\n -r, --remote Connect to a remote extension via relay. Accepts the connector URL from extension Settings (https://relay.api.vibebrowser.app/mcp/), a bare extension UUID, or a ws(s) relay URL. This value is the sole bearer credential — treat it like a password; regenerate it in extension Settings if exposed.\n --devtools Drive your real running Chrome directly over the DevTools Protocol (bypasses the extension relay)\n\n# MCP server tool\nset_remote { \"url\": \"https://relay.api.vibebrowser.app/mcp/\" }\n```\n\nThe `set_remote` MCP server tool hot-reconnects the running MCP server to a different remote relay. It is an MCP tool, not a browser CLI subcommand. It accepts the same three forms as `--remote` (connector URL preferred, bare UUID, or `wss://` URL). The value is the only credential — never place it in logs shared with untrusted parties; regenerate it in extension Settings if exposed.\n\n```bash\n# OpenClaw helper\nVIBE_REMOTE_URL=\"https://relay.api.vibebrowser.app/mcp/00000000-0000-0000-0000-000000000000\"\nPUBLIC_MCP_URL=\"https://browser-bridge.example.com/mcp\"\nnpx -y @vibebrowser/mcp@latest openclaw --remote \"$VIBE_REMOTE_URL\" --public-url \"$PUBLIC_MCP_URL\"\n\n# OpenClaw-compatible browser CLI\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" status\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" status --wait-for-extension --wait-timeout 10000\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" tabs\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" snapshot --json\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" click 12\nnpx -y @vibebrowser/cli@latest --remote \"$VIBE_REMOTE_URL\" type 23 \"hello\" --submit\nnpx -y @vibebrowser/cli@latest --devtools tabs\n\n# Local LLM server\nMODEL=\"qwen3.5\"\nnpx -y @vibebrowser/mcp@latest serve \"$MODEL\"\n -p, --port Ollama API port (default: 11434)\n -y, --yes Skip confirmation prompts\n -d, --debug Enable debug logging\n```\n\n## Troubleshooting\n\n### \"No connection to Vibe extension\"\n\n1. Ensure the Vibe extension is installed in Chrome\n2. Click the extension icon and enable \"MCP External Control\" in Settings\n3. Check that no firewall is blocking localhost connections\n\n### \"OpenClaw cannot reach my local browser bridge\"\n\n1. Start `vibebrowser-mcp` in HTTP mode instead of stdio\n2. Make sure the bridge process is still running on the user's machine\n3. Confirm the extension is in `Remote` mode and connected\n4. Verify the MCP URL in OpenClaw matches the bridge URL. If OpenClaw is cloud-hosted, do not use `127.0.0.1`; use `openclaw --public-url` with a reachable host.\n\n### Debug mode\n\nEnable debug logging to diagnose issues:\n\n```json\n{\n \"mcpServers\": {\n \"vibe\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@vibebrowser/mcp\", \"--debug\"]\n }\n }\n}\n```\n\n## Development\n\n```bash\ngit clone https://github.com/VibeTechnologies/vibe-mcp.git\ncd vibe-mcp\nnpm install\nnpm run build\nnode dist/cli.js --debug\n```\n\n## Keywords\n\nbrowser automation, mcp server, model context protocol, ai browser control, claude desktop browser, cursor browser automation, web automation, chrome automation, ai agent browser, multi-agent browser control, playwright alternative, puppeteer alternative, browser mcp, web scraping ai, ai web agent\n\n## License\n\nApache-2.0\n\n## Links\n\n- [Vibe AI Browser](https://vibebrowser.app) - Main product\n- [Documentation](https://docs.vibebrowser.app) - Full docs\n- [Chrome Extension](https://chromewebstore.google.com/detail/vibe-ai-web-agent/ajfjlohdpfgngdjfafhhcnpmijbbdgln) - Install extension\n- [GitHub Issues](https://github.com/VibeTechnologies/vibe-mcp/issues) - Report bugs\n- [npm Package](https://www.npmjs.com/package/@vibebrowser/mcp) - npm registry\n","changelogMarkdown","videoKey","requiredSecretsJson","publishedAt","priceCents","trialDays",0,"iconType","image","upstream",{"_115":116,"_117":118,"_119":120,"_121":122,"_123":124,"_125":126},"requiredSecrets",[],"tags",[100,101,102,103,104,105,106,107,108,109,110,111,112,113,114],"assets",[],"stats",{"_98":85,"_99":85},"ratingAvg","reviewCount","ai-agent","browser-automation","browser-mcp","chrome","chrome-automation","chrome-devtools-protocol","chrome-extension","claude-desktop","cursor","logged-in-browser","multi-agent","opencode","playwright-alternative","remote-browser","typescript","topics",[100,101,102,103,104,105,106,107,108,109,132,133,134,110,111,112,113,114,135,136],"license","Apache-2.0","language","JavaScript","registryName","io.github.VibeTechnologies/vibe-mcp","homepage","https://vibebrowser.app","packages",[127],{"_128":129,"_130":131,"_75":76},"registryType","npm","identifier","@vibebrowser/mcp","mcp","mcp-server","model-context-protocol","web-automation","windsurf","ENV",{"_139":140,"_141":142,"_143":144},"ENVIRONMENT","production","TURNSTILE_SITE_KEY","0x4AAAAAAB8-rEI2gNDf_QhU","SENTRY_DSN","https://45fd1229f78089d5c3ee78be128030e1@o4510583261757440.ingest.de.sentry.io/4510584253448272"]

Vibe MCP — Real Chrome, Remotely