# Remote MCP Tools
Integrate Model Context Protocol (MCP) tools with the xAI API for extended functionality.
## Overview
Remote MCP Tools allow you to connect external MCP servers to Grok, enabling access to custom tools and data sources through the standardized MCP protocol.
## What is MCP?
The Model Context Protocol (MCP) is an open standard for connecting AI models to external tools and data sources. It provides:
- Standardized tool definitions
- Secure communication
- Extensible architecture
## Configuring Remote MCP Tools
### Basic Setup
```python
from openai import OpenAI
client = OpenAI(
api_key=os.environ.get("XAI_API_KEY"),
base_url="https://api.x.ai/v1"
)
response = client.chat.completions.create(
model="grok-4",
messages=[
{"role": "user", "content": "Query the database for recent orders"}
],
tools=[
{
"type": "mcp",
"mcp": {
"server_url": "https://your-mcp-server.com",
"server_name": "database-tools"
}
}
]
)
```
### Authentication
```python
tools=[
{
"type": "mcp",
"mcp": {
"server_url": "https://your-mcp-server.com",
"server_name": "secure-tools",
"auth": {
"type": "bearer",
"token": "your-auth-token"
}
}
}
]
```
## MCP Server Requirements
Your MCP server must:
1. Implement the MCP protocol specification
2. Be accessible via HTTPS
3. Respond to tool discovery requests
4. Handle tool execution requests
## Tool Discovery
When you specify an MCP server, Grok will:
1. Query the server for available tools
2. Receive tool definitions (name, description, parameters)
3. Make tools available for the conversation
## Example MCP Server Response
```json
{
"tools": [
{
"name": "query_database",
"description": "Execute a read-only SQL query",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "SQL query to execute"
}
},
"required": ["query"]
}
}
]
}
```
## Use Cases
### Database Integration
Connect to your database through an MCP server for natural language queries.
### CRM Access
Query customer data, update records, create tickets.
### Custom APIs
Wrap any internal API as MCP tools for Grok access.
### Document Systems
Search and retrieve from internal document management systems.
## Combining with Built-in Tools
```python
response = client.chat.completions.create(
model="grok-4",
messages=[
{"role": "user", "content": "Search the web for market trends and update our database"}
],
tools=[
{"type": "web_search"},
{
"type": "mcp",
"mcp": {
"server_url": "https://your-mcp-server.com",
"server_name": "database-tools"
}
}
]
)
```
## Security Considerations
1. **Use HTTPS**: Always use encrypted connections
2. **Authenticate**: Implement proper authentication
3. **Limit permissions**: Only expose necessary operations
4. **Audit logging**: Log all tool invocations
5. **Rate limiting**: Protect against abuse
## Error Handling
```python
try:
response = client.chat.completions.create(
model="grok-4",
messages=[{"role": "user", "content": "..."}],
tools=[{"type": "mcp", "mcp": {...}}]
)
except Exception as e:
if "mcp_server_unreachable" in str(e):
print("MCP server is not accessible")
elif "mcp_auth_failed" in str(e):
print("MCP authentication failed")
```
## Best Practices
1. **Descriptive tool names**: Help Grok understand when to use each tool
2. **Clear descriptions**: Document what each tool does
3. **Validate inputs**: Check parameters before execution
4. **Handle timeouts**: MCP servers should respond promptly
5. **Return useful errors**: Help debug issues with clear error messages
