kdoooh000h/gemini-daily-mcp-compressed

Gemini Daily MCP Server (Compressed) ⚡

Token-optimized MCP server for Google Gemini CLI with 43.8% token savings through Progressive Disclosure strategy.

🎯 Why Compressed Version?

This is an experimental token-optimized variant of gemini-daily-mcp that reduces tool schema overhead by 43.8% while maintaining full functionality.

Token Savings Comparison

Impact:

• Per conversation: Save 134 tokens
• 3,000 conversations/month: Save ~$1.61/month
• Scale to 50 tools: Save ~$66/month (61%)

💡 How It Works: Progressive Disclosure

Traditional Approach (Verbose Schema)

{
  "name": "gemini",
  "description": "Execute Gemini CLI in non-interactive mode for AI assistance with high Token efficiency",
  "inputSchema": {
    "properties": {
      "prompt": {
        "type": "string",
        "description": "The coding task, question, or analysis request"
      },
      "sessionId": {
        "type": "string",
        "description": "Optional session ID for conversational context (saves up to 74% tokens in multi-turn conversations)"
      }
    }
  }
}

Size: 731 bytes ≈ 182 tokens

Progressive Disclosure (Compressed Schema)

{
  "name": "gemini",
  "description": "Execute Gemini CLI. Use help for details",
  "inputSchema": {
    "properties": {
      "prompt": {"type": "string"},
      "sessionId": {"type": "string"}
    }
  }
}

Size: 281 bytes ≈ 70 tokens (61.6% savings)

Where Did the Details Go?

They're in the enhanced help tool! When Claude needs details:

# Claude calls: mcp__gemini-daily-compressed__help
# Returns comprehensive documentation:

# Gemini Daily MCP - Detailed Documentation

## gemini Tool

Execute Gemini CLI in non-interactive mode for AI assistance.

**Parameters:**
- prompt (required, string): The coding task, question, or analysis request
- sessionId (optional, string): Session ID for conversational context
  - Saves up to 74% tokens in multi-turn conversations
  - Obtained from previous gemini call's _meta.sessionId
- resetSession (optional, boolean): Reset session history before processing
- model (optional, string): Specify which Gemini model to use
  - Default: gemini-2.5-flash
  - Options: gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.5-pro, gemini-3-pro-preview

**Example:**
{
  "prompt": "Write a Python function to calculate fibonacci",
  "sessionId": "session-abc123",
  "model": "gemini-2.5-flash"
}
...

Result:

• ✅ 90% of conversations: Don't need details → Save tokens
• ✅ 10% of conversations: Call help once → Get full docs
• ✅ Overall: 43.8% token savings

🚀 Features

Same Functionality as Original

• ✅ Full Gemini CLI integration
• ✅ Session management (saves up to 74% tokens in multi-turn conversations)
• ✅ Gemini 3 support (gemini-3-pro-preview with fallback)
• ✅ Multi-model selection (Flash, Flash-Lite, Pro, Gemini 3)
• ✅ Comprehensive error handling

Token Optimization

• ⚡ 43.8% smaller tool schemas
• ⚡ Progressive Disclosure via enhanced help tool
• ⚡ Preserves all type information
• ⚡ Maintains parameter validation

📦 Installation

Prerequisites

• Google Gemini CLI must be pre-installed and configured
  • Install: Follow Google's official Gemini CLI installation guide
  • Setup: Run gemini auth login with your Google account
  • API Key: Set GOOGLE_API_KEY environment variable or configure via gemini config
• Claude Code or Claude Desktop installed

Claude Code

# Clone this repository
git clone https://github.com/kdoooh000h/gemini-daily-mcp-compressed.git
cd gemini-daily-mcp-compressed

# Install dependencies
npm install

# Build
npm run build

# Add to Claude Code
# Edit your .mcp.json:
{
  "mcpServers": {
    "gemini-daily-compressed": {
      "command": "node",
      "args": ["/path/to/gemini-daily-mcp-compressed/dist/index.js"],
      "env": {
        "GEMINI_ENABLED": "true",
        "SESSION_TTL_HOURS": "24",
        "MAX_SESSIONS": "100",
        "GEMINI_CLI_COMMAND": "gemini",
        "LOG_LEVEL": "info"
      }
    }
  }
}

Claude Desktop

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "gemini-daily-compressed": {
      "command": "node",
      "args": ["/path/to/gemini-daily-mcp-compressed/dist/index.js"],
      "env": {
        "GEMINI_ENABLED": "true",
        "SESSION_TTL_HOURS": "24",
        "GEMINI_CLI_COMMAND": "gemini"
      }
    }
  }
}

🔧 Tools

gemini - AI Coding Assistant

Execute Gemini CLI with optional session support.

Basic Usage:

Use gemini to explain this Python function

Advanced Usage:

# With session for multi-turn conversation
Use gemini with sessionId "my-session" to analyze this code

# Continue conversation
Use gemini with sessionId "my-session" to implement the improvements

# Specify model
Use gemini with model "gemini-3-pro-preview" for complex analysis

# Reset session
Use gemini with sessionId "my-session" and resetSession true to start fresh

listSessions - Session Management

List all active conversation sessions.

ping - Connection Test

Test if the MCP server is working properly.

help - Enhanced Documentation

Get detailed documentation for all tools (this is where the detailed info lives in the compressed version).

📊 Compression Details

Strategy

1. Remove verbose descriptions from tool schemas
2. Preserve all functionality:
   • Parameter names
   • Type information
   • Required fields
   • Enum values
3. Enhance help tool to provide complete documentation

Token Breakdown

Trade-offs

Pros ✅:

• Significant token savings (43.8%)
• Lower costs ($66/month for 3K conversations with 50 tools)
• Full functionality preserved
• Claude understands compressed schemas

Cons ⚠️:

• ~10% of conversations may need an extra help call
• Schema less readable for developers
• Need to maintain documentation in two places (schema + help tool)

📈 When to Use Compressed Version?

Recommended For ✅

• High-frequency usage (>100 conversations/day)
• Many tools (>20 tools in your MCP setup)
• Cost-sensitive projects
• Complex tool schemas with verbose descriptions

Use Original Version Instead

• Low-frequency usage (<100 conversations/day)
• Few tools (<10 tools)
• Need maximum schema readability
• Occasional use

🧪 Experimental Status

This is an experimental project to validate the Progressive Disclosure approach for MCP tool schemas.

Testing Results (from lab-02-testbed):

• ✅ All tools function normally
• ✅ help tool provides complete documentation
• ✅ Claude understands compressed schemas
• ✅ 43.8% token savings confirmed
• ⚠️ ~10% scenarios require help call (adds 1 tool invocation)

Feedback Welcome: Please open issues if you encounter any problems or have suggestions.

🔗 Related Projects

• Original Version: gemini-daily-mcp - Full verbose schemas
• Token Analysis: See TOOL-SCHEMA-COMPRESSION-EXPERIMENT.md
• Gemini CLI: Google Gemini CLI

📚 Documentation

• : Gemini model compatibility and selection guide
• : How this was adapted from codex-mcp-server
• API Reference: Use the help tool within Claude for detailed documentation

🛠️ Development

# Install dependencies
npm install

# Development mode
npm run dev

# Build
npm run build

# Test
npm test

# Lint
npm run lint

📝 Version Compatibility

This MCP server is compatible with:

• Gemini CLI: v0.17.0+ (with Gemini 3 support)
• Claude Code: Latest version
• Claude Desktop: Latest version
• MCP SDK: @modelcontextprotocol/sdk ^1.17.3

🤝 Contributing

This is an experimental project. Contributions, issues, and feature requests are welcome!

📄 License

ISC

💬 Questions?

• Use the help tool: mcp__gemini-daily-compressed__help
• Open an issue: GitHub Issues
• Original project: gemini-daily-mcp

⚡ Compressed. Efficient. Same Power.