grok-mcp

Grok MCP Plugin

A Claude Code plugin that integrates xAI's Grok models for multi-model collaboration.

Features

• Natural language triggers: "Ask Grok...", "What does Grok think..."
• Direct commands: /query, /review, /debug, /design for quick Grok queries
• Model selection: Choose from fast, smartest, code-optimized, or reasoning models
• Model-aware timeouts: 90s for slow grok-4, 30s for fast models
• Smart streaming: Auto-enables streaming for complex queries
• Cost tracking: See token usage and cost estimates for every query
• Response caching: Reduce costs with intelligent caching

Installation

Prerequisites

1. xAI API Key: Get one from console.x.ai
2. Claude Code: Version with plugin support

Install from GitHub

# Add the plugin marketplace
/plugin marketplace add mortenbrudvik/ClaudeGrokMcp

# Install the plugin
/plugin install grok-mcp@grok-mcp-dev

# Restart Claude Code

Set API Key

Add to your environment or Claude Code settings:

macOS/Linux (bash/zsh):

export XAI_API_KEY=xai-your-key-here

Windows PowerShell:

$env:XAI_API_KEY = "xai-your-key-here"

Windows Command Prompt:

set XAI_API_KEY=xai-your-key-here

Or configure in Claude Code (works on all platforms):

/config set PLUGIN_ENV_XAI_API_KEY=xai-your-key-here

Usage

Natural Language

Just ask Claude to use Grok:

"Ask Grok what the best practices are for error handling in TypeScript"

"Have Grok analyze this function for bugs"

"What does Grok think about this architecture?"

/grok Command

# Simple query
/grok What is the time complexity of merge sort?

# Specify model
/grok --model fast Explain what a mutex is

# Code analysis
/grok --model code Review this function: function add(a,b) { return a-b }

# Extended reasoning
/grok --model reasoning Think through microservices vs monolith trade-offs

Model Aliases

Alias	Model	Best For	Pricing (per 1M tokens)
auto	grok-4	General queries	$3.00 / $15.00
fast	grok-4-fast	Quick responses	$0.20 / $0.50
smartest	grok-4	Complex analysis	$3.00 / $15.00
code	grok-code-fast-1	Programming tasks	$0.20 / $1.50
reasoning	grok-4-1-fast-reasoning	Multi-step thinking (2M context)	$0.20 / $0.50
cheap	grok-4-fast	Budget-conscious	$0.20 / $0.50
vision	grok-4	Image/vision analysis	$3.00 / $15.00

MCP Tools

grok_query

Query Grok with a question or prompt. Supports vision/image analysis.

{
  query: string,           // Required: The question to ask
  model?: string,          // Model alias or ID (default: "auto")
  context?: string,        // System context to guide response
  max_tokens?: number,     // Max response tokens (default: 4096)
  temperature?: number,    // Sampling temperature 0-2 (default: 0.7)
  image_url?: string,      // Image URL or base64 data URI for vision
  image_detail?: string    // Detail level: "auto", "low", "high"
}

grok_analyze_code

Analyze code for bugs, performance issues, security vulnerabilities, and style problems.

{
  code: string,            // Required: The code to analyze
  language?: string,       // Programming language (auto-detected)
  analysis_type?: string,  // "performance", "bugs", "security", "style", "all"
  context?: string,        // Additional context about the code
  model?: string           // Model to use (default: grok-code-fast-1)
}

grok_reason

Perform extended reasoning and deep thinking on complex problems.

{
  query: string,           // Required: The problem to reason through
  effort?: string,         // "low", "medium", "high" (default: "medium")
  show_thinking?: boolean, // Include reasoning trace (default: true)
  context?: string,        // Additional context
  model?: string           // Model to use (default: grok-4-1-fast-reasoning)
}

grok_estimate_cost

Estimate the cost of a Grok query before execution.

{
  query: string,           // Required: The query to estimate
  model?: string,          // Model to use (default: auto)
  context?: string,        // Additional context
  max_tokens?: number      // Expected output tokens
}

grok_execute_code

Execute Python code server-side for calculations and testing.

{
  code: string,            // Required: Python code to execute
  description?: string,    // What the code should accomplish
  include_output?: boolean,// Include stdout/stderr (default: true)
  max_turns?: number,      // Max iterations 1-10 (default: 3)
  model?: string           // Model to use (default: grok-4-1-fast)
}

grok_search_x

Search X/Twitter and web using Grok agentic search.