Use MCP tools without burning context budget

Quick Start

/use-mcp Get Google Analytics traffic for last 30 days

What happens:

1. Executes via Gemini CLI (preserves Claude context)
2. Gemini CLI auto-connects to MCP servers
3. Discovers available tools
4. Executes MCP tool
5. Returns structured JSON result

Output: JSON response from MCP tool

Why Gemini CLI?

Context Preservation

• MCP servers consume large context
• Running MCP via Claude burns your token budget
• Gemini CLI runs externally, preserves Claude tokens
• Claude stays focused on your task

How It Works

You → Claude → Gemini CLI → MCP Servers → Tools
            ↓
        Executes externally
        Returns JSON only

Syntax

/use-mcp [task]

Execution Method

Uses stdin piping (correct):

echo "$ARGUMENTS. Return JSON only." | gemini -y -m gemini-2.5-flash

DO NOT use -p flag (deprecated, skips MCP):

# WRONG - skips MCP server connections
gemini -y -m gemini-2.5-flash -p "..."

Examples

Example 1: Google Analytics

Input:

/use-mcp Get traffic data for last 30 days

Process:

Executing via Gemini CLI...

MCP Server: google-analytics
Tool: get_traffic_report
Parameters: { period: "30d" }

Result:
{
  "server": "google-analytics",
  "tool": "get_traffic_report",
  "success": true,
  "result": {
    "sessions": 47320,
    "users": 32180,
    "pageviews": 156890,
    "bounce_rate": 42.3,
    "avg_session_duration": "3:24"
  },
  "error": null
}

Summary:
- Sessions: 47,320
- Users: 32,180
- Pageviews: 156,890
- Bounce rate: 42.3%

Example 2: Search Console

Input:

/use-mcp Get top 10 keywords from Search Console

Output:

MCP Server: google-search-console
Tool: get_keywords

Result:
{
  "keywords": [
    { "keyword": "marketing automation", "impressions": 12400, "clicks": 847, "position": 3.2 },
    { "keyword": "email marketing tools", "impressions": 8900, "clicks": 623, "position": 4.1 },
    ...
  ]
}

Example 3: Discord Posting

Input:

/use-mcp Post campaign announcement to #marketing channel

Output:

MCP Server: discord
Tool: send_message

Result:
{
  "success": true,
  "message_id": "123456789",
  "channel": "#marketing",
  "timestamp": "2025-01-29T14:30:00Z"
}

Posted to Discord #marketing channel
Message ID: 123456789

MCP Server Support

Built-in MCP Servers

Configure MCP Servers

MCP servers configured in GEMINI.md (auto-loaded):

{
  "mcpServers": {
    "google-analytics": {
      "command": "npx",
      "args": ["-y", "@google/mcp-server-ga4"]
    }
  }
}

Fallback: mcp-manager Subagent

If Gemini CLI unavailable:

Gemini CLI not found. Using mcp-manager subagent...

✓ mcp-manager: Discovering MCP tools...
✓ Found 23 tools across 4 servers
✓ Executing: google-analytics.get_traffic_report

[Same result as Gemini CLI]

Note: Subagent consumes Claude context. Gemini CLI preferred.

Response Format

All responses are structured JSON:

{
  "server": "server-name",
  "tool": "tool-name",
  "success": true,
  "result": { ... },
  "error": null
}

Flags

• -y: Auto-approve tool execution (no prompts)
• -m gemini-2.5-flash: Use fast Gemini model

Workflow Integration

Analytics Reporting

/use-mcp Get GA4 traffic for last 30 days
/analyze traffic  # Analyze with AI

SEO Tracking

/use-mcp Get Search Console keyword rankings
/seo keywords "top performing keywords"

Multi-Channel Publishing

/social twitter thread "product launch"
/use-mcp Post to Discord #announcements
/use-mcp Post to Slack #marketing

Troubleshooting

Gemini CLI not found

# Install Gemini CLI
npm install -g @google/generative-ai-cli

# Or use fallback
# /use-mcp automatically uses mcp-manager subagent

MCP server not connected

# Check GEMINI.md configuration
# Ensure MCP server installed
# Restart Gemini CLI

Tool execution failed

# Check MCP server logs
# Verify API keys
# Check tool parameters

Related Commands

• /analyze - Analyze MCP data
• /campaign - Use analytics in campaigns

Related Skills

• mcp-management: MCP server operations
• mcp-builder: MCP tool development

External execution. Zero context cost. Use MCP tools without burning tokens.