grok-build-mcp

Bring xAI's Grok Build CLI into any MCP host as a peer reviewer, adversary, and consultant — alongside whatever main model you're already running.

grok-build-mcp is a small Model Context Protocol server that wraps the grok CLI. Your existing agent (Claude Code, Cursor, Cline, OpenClaw, …) can call into Grok for second-opinion code review, adversarial testing, or extended chat — without leaving its session.

繁體中文版：README.zh-TW.md

What you get

Four tools, all stateless, all stdout-only:

Prerequisites

• Node.js ≥ 18

• The Grok CLI installed and authenticated:

  curl -fsSL https://x.ai/cli/install.sh | bash
  grok  # first run handles auth

Install

npm install -g grok-build-mcp
# or use npx — no install needed
npx grok-build-mcp

Wire it into your MCP host

Claude Code

claude mcp add grok npx -- grok-build-mcp

Or edit ~/.claude.json directly:

{
  "mcpServers": {
    "grok": {
      "command": "npx",
      "args": ["-y", "grok-build-mcp"]
    }
  }
}

Cursor

Create .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "grok": {
      "command": "npx",
      "args": ["-y", "grok-build-mcp"]
    }
  }
}

Cline (VS Code)

Settings → Cline → MCP Servers:

{
  "grok": {
    "command": "npx",
    "args": ["-y", "grok-build-mcp"]
  }
}

Any other MCP host

grok-build-mcp speaks plain stdio MCP. Point any client at npx -y grok-build-mcp and it works.

Tool reference

grok_chat

{ "prompt": "Explain consistent hashing in two sentences." }

Optional: model to override the default Grok model.

grok_review

{ "base_ref": "main", "focus": "security" }

If diff is omitted, runs git diff <base_ref>...HEAD in cwd (defaults to your host's working directory). Returns a markdown review with verdict, per-dimension scores (correctness / readability / architecture / security / performance), and concrete fix-it items.

grok_consult

{
  "messages": [
    { "role": "system", "content": "You are a senior backend engineer." },
    { "role": "user", "content": "How would you cache this query?" },
    { "role": "assistant", "content": "Two options..." },
    { "role": "user", "content": "What's the failure mode of option 2?" }
  ]
}

The server is stateless — the caller passes the full thread each time. Most MCP hosts handle this naturally.

grok_challenge

{
  "code": "function transfer(from, to, amount) { from.balance -= amount; to.balance += amount; }",
  "context": "Node.js, called concurrently from HTTP handlers"
}

Returns severity-ranked issues (Critical / High / Medium / Low) with concrete reproductions and patches.

Configuration

Authentication and model defaults live in the Grok CLI itself (~/.grok/config.toml).

Roadmap

• v0.1 (this release): four stateless tools, stdio transport

• v0.2: server-side session persistence so grok_consult can take a conversation_id

• v0.3: streaming responses through MCP progress notifications

Development

git clone https://github.com/howardpen9/grok-mcp.git
cd grok-mcp
npm install
npm test
npm run build

License

MIT