📈 Request Generation Frequency
Timeline distribution of successful and rate-limited queries
Key Pool & Accounts Management
Add, remove, or monitor your Grok SSO account tokens
➕ Add New Account Token
Tokens are automatically saved to keys.txt💡 Tip: Log in on grok.com, press F12 -> Storage/Application -> Cookies -> copy sso.
👤 Key Pool Accounts 0
Priority: accounts with fewer remaining queries are used first
🤖 Models & Limits
Live rates per model identifier
⚡ Interactive Prompt Console
Sends private incognito requests through the active key pool
Grok Search Agent — Model Context Protocol (MCP)
Connect this real-time web search agent directly to Cursor, Claude Desktop, Claude Code, Windsurf, or custom AI agents via HTTP SSE.
1. In Cursor, open Settings → Features → MCP → Click Add New MCP Server.
2. Set Type to SSE and paste the configuration or URL:
{
"mcpServers": {
"grok-search": {
"url": "http://:/mcp/sse",
"headers": {
"X-API-Key": ""
}
}
}
}
Paste into ~/.config/Claude/claude_desktop_config.json (Mac/Linux) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"grok-search": {
"url": "http://:/mcp/sse",
"headers": {
"X-API-Key": ""
}
}
}
}
⚙️ Dynamic MCP Tool Behavior
When only 1 model is checked in the dashboard, only ask_grok(message) is registered. The model is automatically injected without asking the agent.
When multiple models are checked, list_grok_models() and ask_grok(message, model) with strict Enum validation are registered.
Tool descriptions instruct the calling AI to use Grok strictly for real-time web retrieval, citations, and links, while doing complex logical deductions independently.
💻 Local Stdio Alternative
If you prefer running a direct command subprocess instead of an HTTP connection, run via Python stdio:
{
"mcpServers": {
"grok-search": {
"command": "/path/to/venv/bin/python",
"args": ["/path/to/Grok-MCP/mcp_server.py"],
"env": {
"API_KEY": "grok-secret-key-123"
}
}
}
}