Grok CLI MCP Server
# grok-cli-mcp
[](https://badge.fury.io/py/grok-cli-mcp)
[](https://www.python.org/downloads/)
[](https://opensource.org/licenses/MIT)
> MCP server that wraps the Grok CLI, providing seamless access to Grok AI models through the Model Context Protocol.
## What is this?
`grok-cli-mcp` is a [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server that acts as a bridge between MCP clients (like Claude Code, Cline, Cursor) and the [Grok CLI](https://docs.x.ai/docs). Instead of implementing direct API calls, it leverages the official Grok CLI tool, providing:
- **Three specialized tools**: `grok_query` (general queries), `grok_chat` (multi-turn conversations), `grok_code` (code generation)
- **Simple configuration**: Just install the Grok CLI and set your API key
- **Future-proof**: Automatically benefits from CLI improvements (OAuth, pricing plans, etc.)
- **Minimal maintenance**: No need to track Grok API changes
## Why a CLI Wrapper?
### Benefits
✅ **Leverage existing tooling**: Uses the official Grok CLI, ensuring compatibility and stability
✅ **Future OAuth support**: When Grok CLI adds OAuth authentication, this wrapper will support it automatically without code changes
✅ **Fixed pricing plans**: Can benefit from fixed monthly pricing (like Codex/ChatGPT/Gemini) when Grok introduces CLI-specific plans, rather than paying per API token
✅ **Organization-friendly**: Many organizations prefer audited CLI tools over direct API integrations for security and compliance
✅ **Simpler codebase**: ~400 lines vs 1500+ for a full API client implementation
✅ **Fewer dependencies**: No HTTP client libraries, request/response handling, or complex networking code
✅ **Automatic updates**: CLI bug fixes and new features propagate without code changes
### Tradeoffs
⚠️ **Performance overhead**: Extra process spawning adds ~50-200ms latency per request
⚠️ **CLI dependency**: Requires Grok CLI to be installed and in PATH
⚠️ **Limited control**: Can't access low-level API features not exposed by CLI
⚠️ **Error handling**: CLI error messages may be less structured than API responses
⚠️ **No streaming**: Limited to CLI streaming capabilities (if any)
### When to use this
**Perfect for:**
- Development and prototyping workflows
- Internal tools and automation (<100 req/min)
- Organizations preferring CLI tools over API libraries
- Workflows where convenience matters more than milliseconds
- Teams wanting to benefit from future CLI-specific pricing/features
**Consider direct API for:**
- High-throughput production systems (>1000 req/min)
- Latency-critical applications (<50ms requirements)
- Advanced API features not exposed by CLI
- Streaming response requirements
## Prerequisites
Before installing `grok-cli-mcp`, ensure you have:
1. **Grok CLI**: Install from [X.AI's documentation](https://docs.x.ai/docs)
```bash
# Installation instructions vary by platform
# See https://docs.x.ai/docs for latest instructions
```
2. **Python 3.10+**: Check your version
```bash
python3 --version
```
3. **Grok API Key**: Obtain from [X.AI console](https://console.x.ai/)
## Installation
### Option 1: Install from PyPI (Recommended)
```bash
pip install grok-cli-mcp
```
### Option 2: Install with uv
```bash
uv pip install grok-cli-mcp
```
### Option 3: Install with pipx (isolated environment)
```bash
pipx install grok-cli-mcp
```
### Option 4: Install from source
```bash
git clone https://github.com/BasisSetVentures/grok-cli-mcp.git
cd grok-cli-mcp
pip install -e .
```
### Option 5: Development installation
```bash
git clone https://github.com/BasisSetVentures/grok-cli-mcp.git
cd grok-cli-mcp
pip install -e ".[dev]"
```
## Quick Start
### 1. Set up your environment
```bash
# Required: Set your Grok API key
export GROK_API_KEY="your-api-key-here"
# Optional: Specify custom Grok CLI path
export GROK_CLI_PATH="/custom/path/to/grok"
```
For permanent setup, add to your shell profile (`~/.bashrc`, `~/.zshrc`, etc.):
```bash
echo 'export GROK_API_KEY="your-api-key-here"' >> ~/.bashrc
source ~/.bashrc
```
### 2. Test the server
```bash
# Run the server directly
python -m grok_cli_mcp
# Or use the command
grok-mcp
# Should start and wait for stdin (Ctrl+C to exit)
```
### 3. Configure for MCP clients
#### For Claude Code
Add to your `.mcp.json`:
```json
{
"mcpServers": {
"grok": {
"type": "stdio",
"command": "python",
"args": ["-m", "grok_cli_mcp"],
"env": {
"GROK_API_KEY": "your-api-key-here"
}
}
}
}
```
#### For Cline (VS Code)
Add to `~/.cline/mcp_settings.json`:
```json
{
"mcpServers": {
"grok": {
"command": "python",
"args": ["-m", "grok_cli_mcp"],
"env": {
"GROK_API_KEY": "your-api-key-here"
}
}
}
}
```
#### For Cursor
Add to `~/.cursor/mcp.json`:
```json
{
"grok": {
"command": "python",
"args": ["-m", "grok_cli_mcp"],
"env": {
"GROK_API_KEY": "your-api-key-here"
}
}
}
```
**⚠️ Security Warning**: Never commit API keys to version control. Use environment variables or a secrets manager.
## Usage Examples
### Tool: grok_query
Send a simple prompt to Grok:
```json
{
"tool": "grok_query",
"arguments": {
"prompt": "Explain quantum computing in simple terms",
"model": "grok-code-fast-1",
"timeout_s": 120
}
}
```
**Response**: Plain text answer from Grok
### Tool: grok_chat
Multi-turn conversation with message history:
```json
{
"tool": "grok_chat",
"arguments": {
"messages": [
{"role": "user", "content": "What is MCP?"},
{"role": "assistant", "content": "MCP is Model Context Protocol..."},
{"role": "user", "content": "How does it work?"}
],
"model": "grok-code-fast-1",
"timeout_s": 120
}
}
```
**Response**: Grok's answer considering the conversation history
### Tool: grok_code
Code generation with language hints and context:
```json
{
"tool": "grok_code",
"arguments": {
"task": "Create a Python function to parse JSON with error handling",
"language": "python",
"context": "Using standard library only, no external dependencies",
"timeout_s": 180
}
}
```
**Response**: Complete, usable Python code with explanations
### Advanced: Raw Output Mode
Get structured response with full details:
```json
{
"tool": "grok_query",
"arguments": {
"prompt": "Explain async/await",
"raw_output": true
}
}
```
**Response**:
```json
{
"text": "Async/await is...",
"messages": [{"role": "assistant", "content": "..."}],
"raw": "...",
"model": "grok-code-fast-1"
}
```
## Configuration
### Environment Variables
| Variable | Required | Default | Description |
|----------|----------|---------|-------------|
| `GROK_API_KEY` | **Yes** | - | Your Grok API key from X.AI console |
| `GROK_CLI_PATH` | No | `/opt/homebrew/bin/grok` | Path to Grok CLI binary |
### Model Selection
Available models (as of 2025-12):
- `grok-code-fast-1` - Fast model for code tasks
- `grok-2` - Main model for general tasks
- Other models per [Grok CLI documentation](https://docs.x.ai/docs)
Specify model in each tool call or omit for CLI default.
### Timeout Configuration
Default timeouts by tool:
- `grok_query`: 120 seconds
- `grok_chat`: 120 seconds
- `grok_code`: 180 seconds
Adjust via `timeout_s` parameter for complex tasks.
## Troubleshooting
### "Grok CLI not found"
**Problem**: Server can't locate the Grok CLI binary
**Solutions**:
1. Verify installation:
```bash
which grok
```
2. Set explicit path:
```bash
export GROK_CLI_PATH="/path/to/grok"
```
3. Add to PATH:
```bash
export PATH="$PATH:/opt/homebrew/bin"
```
### "GROK_API_KEY is not set"
**Problem**: API key not in environment
**Solutions**:
1. Export in shell:
```bash
export GROK_API_KEY="xai-..."
```
2. Add to shell profile (`.bashrc`, `.zshrc`):
```bash
echo 'export GROK_API_KEY="xai-..."' >> ~/.zshrc
source ~/.zshrc
```
3. Use `.env` file with python-dotenv (see `examples/.env.example`)
### "Grok CLI timed out"
**Problem**: Request took too long
**Solutions**:
1. Increase timeout:
```json
{"timeout_s": 300}
```
2. Simplify prompt or break into smaller requests
3. Check network connectivity
### JSON parsing errors
**Problem**: CLI output isn't valid JSON
**Solutions**:
1. Update Grok CLI to latest version:
```bash
# Update instructions vary by installation method
```
2. Check for CLI warnings/errors
3. Use `raw_output=true` to see raw CLI response:
```json
{"raw_output": true}
```
### Permission errors
**Problem**: Can't execute Grok CLI
**Solutions**:
1. Make CLI executable:
```bash
chmod +x /path/to/grok
```
2. Check file ownership and permissions
3. Verify CLI works standalone:
```bash
grok -p "test"
```
For more solutions, see [docs/troubleshooting.md](docs/troubleshooting.md).
## Security Best Practices
### Never Commit Secrets
**❌ DO NOT:**
- Commit `.env` files with real API keys
- Include API keys in `.mcp.json` tracked by git
- Share API keys in issues or pull requests
- Hardcode keys in Python files
**✅ DO:**
- Use environment variables: `export GROK_API_KEY="..."`
- Use shell RC files: `~/.bashrc`, `~/.zshrc`
- Use secrets managers in production: AWS Secrets Manager, HashiCorp Vault
- Rotate keys immediately if accidentally exposed
### Obtaining API Keys
1. Visit [X.AI Console](https://console.x.ai/)
2. Sign in with your X.AI account
3. Navigate to API Keys section
4. Generate a new key
5. Store securely (1Password, Bitwarden, etc.)
6. Set as environment variable
### Key Rotation
If you accidentally expose your API key:
1. **Immediately** revoke the key in X.AI console
2. Generate a new key
3. Update environment variables
4. Check git history for exposed keys
5. Consider using tools like `gitleaks` to scan for secrets
### Reporting Security Issues
**Do NOT** open public issues for security vulnerabilities.
Please report security concerns responsibly through GitHub Security Advisories or by contacting the maintainers directly.
## Architecture & Design
This project follows a **CLI wrapper pattern** rather than direct API integration. Key design decisions:
1. **Process isolation**: Each Grok request spawns a subprocess for CLI execution
2. **JSON parsing with fallback**: Attempts structured parsing, falls back to raw output
3. **Context propagation**: Uses FastMCP's Context for logging and progress updates
4. **Async execution**: All operations are async-first for non-blocking behavior
For detailed architecture discussion, see [docs/architecture.md](docs/architecture.md).
## Development
### Running tests
```bash
# Install dev dependencies
pip install -e ".[dev]"
# Run all tests
pytest
# Run with coverage
pytest --cov=grok_cli_mcp --cov-report=html
# Run specific test file
pytest tests/test_utils.py
```
### Code formatting
```bash
# Format code
black .
# Lint code
ruff check --fix .
```
### Type checking
```bash
mypy src/
```
## Contributing
Contributions are welcome! Please:
1. Fork the repository
2. Create a feature branch (`git checkout -b feature/amazing-feature`)
3. Commit your changes (`git commit -m 'Add amazing feature'`)
4. Push to the branch (`git push origin feature/amazing-feature`)
5. Open a Pull Request
**Please ensure**:
- Tests pass (`pytest`)
- Code is formatted (`black`, `ruff`)
- Type hints are correct (`mypy`)
- Documentation is updated
## License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
## Acknowledgments
- Built with [FastMCP](https://github.com/jlowin/fastmcp) by Jeremiah Lowin
- Uses [Model Context Protocol](https://modelcontextprotocol.io) SDK by Anthropic
- Wraps [Grok CLI](https://docs.x.ai/docs) from X.AI
## Support
- **Documentation**: [README](README.md) • [Architecture](docs/architecture.md) • [Troubleshooting](docs/troubleshooting.md)
- **Issues**: [GitHub Issues](https://github.com/BasisSetVentures/grok-cli-mcp/issues)
- **Discussions**: [GitHub Discussions](https://github.com/BasisSetVentures/grok-cli-mcp/discussions)
- **Grok Documentation**: [docs.x.ai](https://docs.x.ai/docs)
---
**Made by [Basis Set Ventures](https://github.com/BasisSetVentures)** with Claude Code and FastMCP
TDQS
Scored across 3 tools
The tools have distinct primary purposes: grok_chat handles multi-turn conversations by flattening messages, grok_code focuses on code-related queries with language hints, and grok_query is for single-prompt interactions with raw output options. However, grok_chat and grok_query both involve sending prompts to Grok, which could cause minor confusion about when to use each for simple queries.
All tool names follow a consistent 'grok_' prefix pattern with descriptive suffixes (chat, code, query), using snake_case uniformly. This makes the tools easily identifiable and predictable within the server's domain.
With 3 tools, the count is reasonable for a CLI server focused on interacting with Grok, covering chat, code, and general query use cases. It is slightly lean but not insufficient, as each tool addresses a specific aspect of the Grok interface.
The tools cover core functionalities for querying Grok (chat, code, general queries), but there are notable gaps such as missing operations for managing sessions, handling file uploads, or configuring settings, which might limit advanced workflows in a CLI context.