gemini-mcp（Gemini MCP 搜索）

Gemini MCP Server Management

Comprehensive MCP (Model Context Protocol) server integration for extending Gemini CLI with custom tools and capabilities.

What is MCP?

MCP (Model Context Protocol) allows Gemini to connect to external servers that provide additional tools and capabilities:

• Database connections
• API integrations
• Custom business logic
• Enterprise systems
• Specialized tools

Quick Start

List MCP Servers

# View configured servers
gemini mcp list

# Check server status
gemini -i
/mcp

# Test server connection
gemini mcp test <server-name>

Add MCP Server

# Interactive setup
gemini mcp add

# Manual configuration
gemini mcp add \
  --name "my-server" \
  --command "python" \
  --args "-m my_mcp_server" \
  --cwd "./mcp-servers/"

Remove Server

gemini mcp remove <server-name>

MCP Configuration

Configuration File

// ~/.gemini/mcp-servers.json or ./.gemini/mcp-servers.json
{
  "mcpServers": {
    "database-tools": {
      "command": "python",
      "args": ["-m", "database_mcp_server"],
      "cwd": "./mcp-tools/database",
      "env": {
        "DATABASE_URL": "postgresql://localhost/mydb",
        "DB_PASSWORD": "$DB_PASSWORD_FROM_ENV"
      },
      "timeout": 30000,
      "trust": false
    },
    "api-gateway": {
      "command": "node",
      "args": ["./api-mcp-server.js"],
      "cwd": "./mcp-tools/api",
      "env": {
        "API_KEY": "$API_KEY",
        "BASE_URL": "https://api.example.com"
      },
      "includeTools": ["getUser", "createOrder"],
      "excludeTools": ["deleteUser"]
    },
    "analytics": {
      "command": "docker",
      "args": ["run", "-p", "8080:8080", "analytics-mcp:latest"],
      "timeout": 60000,
      "trust": true
    }
  }
}

Security Settings

{
  "mcpServers": {
    "secure-server": {
      "command": "python",
      "args": ["secure_server.py"],
      "trust": false,  // Require confirmation for tool calls
      "includeTools": ["safe_read", "safe_write"],
      "excludeTools": ["dangerous_delete"],
      "allowedDomains": ["*.internal.com"],
      "maxConcurrent": 5,
      "rateLimit": {
        "requests": 100,
        "window": 60000  // per minute
      }
    }
  }
}

Building MCP Servers

Python MCP Server

# database_mcp_server.py
import json
import sys
import psycopg2
from typing import Dict, Any

class DatabaseMCP:
    def __init__(self):
        self.conn = psycopg2.connect(
            os.environ.get('DATABASE_URL')
        )

    def handle_request(self, request: Dict[str, Any]):
        method = request.get('method')
        params = request.get('params', {})

        if method == 'query':
            return self.execute_query(params.get('sql'))
        elif method == 'insert':
            return self.insert_data(
                params.get('table'),
                params.get('data')
            )

    def execute_query(self, sql: str):
        cursor = self.conn.cursor()
        cursor.execute(sql)
        return cursor.fetchall()

    def get_tools(self):
        return [
            {
                "name": "database_query",
                "description": "Execute SQL query",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "sql": {
                            "type": "string",
                            "description": "SQL query to execute"
                        }
                    },
                    "required": ["sql"]
                }
            },
            {
                "name": "database_insert",
                "description": "Insert data into table",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "table": {"type": "string"},
                        "data": {"type": "object"}
                    },
                    "required": ["table", "data"]
                }
            }
        ]

if __name__ == "__main__":
    server = DatabaseMCP()
    server.start()

Node.js MCP Server

// api-mcp-server.js
const { MCPServer } = require('@modelcontextprotocol/server');
const axios = require('axios');

class APIMCPServer extends MCPServer {
  constructor() {
    super();
    this.baseURL = process.env.BASE_URL;
    this.apiKey = process.env.API_KEY;
  }

  async getTools() {
    return [
      {
        name: 'api_get',
        description: 'Make GET request to API',
        parameters: {
          type: 'object',
          properties: {
            endpoint: {
              type: 'string',
              description: 'API endpoint path'
            },
            params: {
              type: 'object',
              description: 'Query parameters'
            }
          },
          required: ['endpoint']
        }
      },
      {
        name: 'api_post',
        description: 'Make POST request to API',
        parameters: {
          type: 'object',
          properties: {
            endpoint: { type: 'string' },
            body: { type: 'object' }
          },
          required: ['endpoint', 'body']
        }
      }
    ];
  }

  async handleToolCall(name, params) {
    const headers = {
      'Authorization': `Bearer ${this.apiKey}`,
      'Content-Type': 'application/json'
    };

    switch(name) {
      case 'api_get':
        const response = await axios.get(
          `${this.baseURL}${params.endpoint}`,
          { headers, params: params.params }
        );
        return response.data;

      case 'api_post':
        const postResponse = await axios.post(
          `${this.baseURL}${params.endpoint}`,
          params.body,
          { headers }
        );
        return postResponse.data;

      default:
        throw new Error(`Unknown tool: ${name}`);
    }
  }
}

const server = new APIMCPServer();
server.start();

Docker MCP Server

# Dockerfile for MCP server
FROM python:3.11-slim

WORKDIR /app

# Install dependencies
COPY requirements.txt .
RUN pip install -r requirements.txt

# Copy server code
COPY mcp_server.py .

# MCP protocol port
EXPOSE 8080

# Start server
CMD ["python", "mcp_server.py"]

# Build and run
docker build -t my-mcp-server .
docker run -p 8080:8080 -e DATABASE_URL="$DATABASE_URL" my-mcp-server

Usage Patterns

Database Operations

# Configure database MCP
cat > ~/.gemini/mcp-servers.json << 'EOF'
{
  "mcpServers": {
    "postgres": {
      "command": "python",
      "args": ["-m", "postgres_mcp"],
      "env": {
        "DATABASE_URL": "postgresql://user:pass@localhost/db"
      }
    }
  }
}
EOF

# Use with YOLO for automated database operations
gemini --yolo -p "Query the users table and generate comprehensive report of active users from last week"
gemini --yolo -p "Create detailed sales analysis with charts from the orders table"

API Integration

# Configure API MCP
gemini mcp add \
  --name "stripe" \
  --command "node" \
  --args "stripe-mcp.js" \
  --env "STRIPE_KEY=$STRIPE_SECRET_KEY"

# Use with YOLO for automated API operations
gemini --yolo -p "Create a new customer and subscription using Stripe with full setup"
gemini --yolo -p "Generate comprehensive transaction report with analytics for this month"

Custom Business Logic

# Configure business logic MCP
gemini mcp add \
  --name "business-rules" \
  --command "java" \
  --args "-jar business-mcp.jar"

# Apply business rules with YOLO automation
gemini --yolo -p "Validate this order against all business rules and generate compliance report"
gemini --yolo -p "Calculate pricing using custom algorithm and create detailed breakdown"

YOLO Mode with MCP Servers

YOLO mode with MCP servers enables powerful automation for trusted operations:

Automated Data Operations

# Database automation with YOLO
gemini --yolo -p "Using the database MCP:
1. Query all user activity from last 30 days
2. Generate engagement analytics
3. Create user segmentation report
4. Export findings to CSV
5. Send summary email to stakeholders"

# Multi-database operations
gemini --yolo -p "Synchronize data between PostgreSQL and Redis:
1. Read user sessions from Redis
2. Update last_active in PostgreSQL
3. Clean expired sessions
4. Generate sync report"

API Workflow Automation

# Automated API orchestration
gemini --yolo -p "Using Stripe and database MCP servers:
1. Fetch all subscription cancellations from Stripe
2. Update user status in our database
3. Send personalized retention offers
4. Log all actions for audit
5. Generate retention campaign report"

# Multi-service integration
gemini --yolo -p "Complete order processing workflow:
1. Validate order via business rules MCP
2. Process payment via Stripe MCP
3. Update inventory via database MCP
4. Send confirmation via email MCP
5. Log transaction and generate receipt"

Enterprise Automation

# Automated compliance reporting
gemini --yolo -p "Generate quarterly compliance report:
1. Query all financial data via database MCP
2. Validate against regulations via compliance MCP
3. Generate charts and visualizations
4. Create executive summary
5. Export to PDF and store securely"

# Infrastructure monitoring
gemini --yolo -p "Complete infrastructure health check:
1. Query metrics from monitoring MCP
2. Check service status via K8s MCP
3. Analyze logs via logging MCP
4. Generate incident reports
5. Update status dashboard"

Safe YOLO Practices for MCP

# ✅ SAFE with YOLO (read-only or low-risk)
gemini --yolo -p "Generate analytics dashboard from database MCP"
gemini --yolo -p "Sync read-only data between MCP services"
gemini --yolo -p "Create comprehensive status reports from all MCPs"

# ⚠️ USE CAUTION (write operations)
# Review first, then use YOLO if confident
gemini -p "Plan user data migration between databases"
# After review:
gemini --yolo -p "Execute the reviewed migration plan"

# 🔒 NEVER YOLO (critical operations)
# Always manual approval for:
# - Production data deletion
# - Security configuration changes
# - Financial transactions above thresholds

MCP Server Management Automation

#!/bin/bash
# Automated MCP server lifecycle management

manage_mcp_servers() {
  local operation="$1"

  case $operation in
    health-check)
      gemini --yolo -p "Check health of all MCP servers and create status report"
      ;;

    restart-all)
      gemini --yolo -p "Safely restart all MCP servers in dependency order"
      ;;

    update-configs)
      gemini --yolo -p "Update all MCP server configurations and validate"
      ;;

    deploy-new)
      gemini --yolo -p "Deploy new MCP servers from configs and test connections"
      ;;
  esac
}

# Usage
manage_mcp_servers health-check

Advanced Workflows

Multi-Server Orchestration

#!/bin/bash
# Orchestrate multiple MCP servers

orchestrate_mcp() {
  # Start all servers
  gemini mcp start database
  gemini mcp start api
  gemini mcp start analytics

  # Execute workflow
  gemini --yolo -p "Using all available MCP tools:
  1. Query user data from database
  2. Enrich with API data
  3. Analyze with analytics tools
  4. Generate comprehensive report"

  # Stop servers
  gemini mcp stop --all
}

Dynamic Server Management

#!/bin/bash
# Dynamically add servers based on project

setup_project_mcp() {
  local project_type="$1"

  case $project_type in
    ecommerce)
      gemini mcp add --name "payment" --command "payment-mcp"
      gemini mcp add --name "inventory" --command "inventory-mcp"
      gemini mcp add --name "shipping" --command "shipping-mcp"
      ;;

    analytics)
      gemini mcp add --name "bigquery" --command "bq-mcp"
      gemini mcp add --name "tableau" --command "tableau-mcp"
      ;;

    devops)
      gemini mcp add --name "kubernetes" --command "k8s-mcp"
      gemini mcp add --name "terraform" --command "tf-mcp"
      ;;
  esac

  echo "MCP servers configured for $project_type project"
}

Health Monitoring

#!/bin/bash
# Monitor MCP server health

monitor_mcp_health() {
  while true; do
    echo "=== MCP Server Status ==="

    for server in $(gemini mcp list --json | jq -r '.servers[].name'); do
      if gemini mcp test "$server" > /dev/null 2>&1; then
        echo "✓ $server: Healthy"
      else
        echo "✗ $server: Unhealthy"

        # Attempt restart
        echo "  Attempting restart..."
        gemini mcp restart "$server"
      fi
    done

    sleep 30
  done
}

Load Balancing

// Load-balanced MCP configuration
{
  "mcpServers": {
    "api-pool": {
      "type": "pool",
      "strategy": "round-robin",  // or "random", "least-connections"
      "servers": [
        {
          "command": "node",
          "args": ["api-1.js"],
          "port": 8081
        },
        {
          "command": "node",
          "args": ["api-2.js"],
          "port": 8082
        },
        {
          "command": "node",
          "args": ["api-3.js"],
          "port": 8083
        }
      ],
      "healthCheck": {
        "interval": 10000,
        "timeout": 5000,
        "path": "/health"
      }
    }
  }
}

Security Best Practices

Authentication

{
  "mcpServers": {
    "secure-server": {
      "command": "python",
      "args": ["server.py"],
      "auth": {
        "type": "bearer",
        "token": "$MCP_AUTH_TOKEN"
      },
      "tls": {
        "enabled": true,
        "cert": "/path/to/cert.pem",
        "key": "/path/to/key.pem",
        "ca": "/path/to/ca.pem"
      }
    }
  }
}

Tool Restrictions

#!/bin/bash
# Restrict MCP tool access

restrict_mcp_tools() {
  local server="$1"
  local allowed_tools=("$@")

  # Generate config with restrictions
  cat > ~/.gemini/mcp-restrictions.json << EOF
{
  "$server": {
    "includeTools": [${allowed_tools[@]}],
    "requireConfirmation": true,
    "logAllCalls": true,
    "maxCallsPerMinute": 10
  }
}
EOF

  # Apply restrictions
  gemini mcp update "$server" --config ~/.gemini/mcp-restrictions.json
}

# Usage
restrict_mcp_tools "database" "read_only_query" "get_schema"

Audit Logging

# Auditing MCP server
import logging
import json
from datetime import datetime

class AuditedMCPServer:
    def __init__(self):
        self.audit_log = logging.getLogger('mcp.audit')
        self.audit_log.setLevel(logging.INFO)

        # Setup audit file
        handler = logging.FileHandler('/var/log/mcp-audit.log')
        formatter = logging.Formatter(
            '%(asctime)s - %(name)s - %(message)s'
        )
        handler.setFormatter(formatter)
        self.audit_log.addHandler(handler)

    def handle_tool_call(self, tool, params, user_context):
        # Log before execution
        self.audit_log.info(json.dumps({
            'event': 'tool_call_start',
            'tool': tool,
            'params': params,
            'user': user_context,
            'timestamp': datetime.utcnow().isoformat()
        }))

        try:
            result = self.execute_tool(tool, params)

            # Log success
            self.audit_log.info(json.dumps({
                'event': 'tool_call_success',
                'tool': tool,
                'result_size': len(str(result))
            }))

            return result

        except Exception as e:
            # Log failure
            self.audit_log.error(json.dumps({
                'event': 'tool_call_failure',
                'tool': tool,
                'error': str(e)
            }))
            raise

Troubleshooting

Common Issues

1. Server Won't Start

# Check logs
gemini mcp logs <server-name>

# Test manually
python -m my_mcp_server --debug

# Check port conflicts
lsof -i :8080

1. Connection Timeout

# Increase timeout
gemini mcp update <server> --timeout 60000

# Check network
ping localhost
telnet localhost 8080

1. Tool Not Available

# List available tools
gemini -i
/tools

# Refresh tool list
gemini mcp refresh <server>

# Check server implementation
gemini mcp debug <server>

Debug Mode

# Enable debug logging
export GEMINI_MCP_DEBUG=true

# Verbose output
gemini --verbose mcp test <server>

# Trace tool calls
gemini --trace -p "Use MCP tools to query database"

Performance Optimization

Connection Pooling

{
  "mcpServers": {
    "database": {
      "command": "python",
      "args": ["db_mcp.py"],
      "pool": {
        "min": 2,
        "max": 10,
        "idle": 300000  // 5 minutes
      }
    }
  }
}

Caching

# Enable MCP response caching
export GEMINI_MCP_CACHE=true
export GEMINI_MCP_CACHE_TTL=300  # seconds

# Clear cache
gemini mcp cache clear

Related Skills

• gemini-cli: Main Gemini CLI integration
• gemini-auth: Authentication management
• gemini-chat: Interactive chat sessions
• gemini-tools: Tool execution workflows