MCP Playground

Research and analysis toolkit with real-time data access via Model Context Protocol (MCP).

Live Server: https://mcp-playground.fastmcp.app/mcp

Quick Start

git clone <your-repo>
cd mcp_arena
uv sync --extra local
uv run streamlit run app.py

MCP Client Setup

Claude Desktop

Option 1: Remote Server (Recommended) Add to claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-playground": {
      "type": "http",
      "url": "https://mcp-playground.fastmcp.app/mcp"
    }
  }
}

Option 2: Local Development Add to claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-playground-local": {
      "command": "uv",
      "args": ["run", "python", "mcp_server.py"]
    }
  }
}

Note: Run this command from your mcp_arena directory

Claude Code

1. Cmd/Ctrl + Shift + P → "MCP: Add Server"
2. Choose "Add remote HTTP server"
3. URL: https://mcp-playground.fastmcp.app/mcp

LMStudio (v0.3.17+)

1. Program tab → "Edit mcp.json"
2. Add configuration:

{
  "mcp-arena": {
    "url": "https://mcp-playground.fastmcp.app/mcp",
  }
}

Available Tools

Research & Analysis

• web_search(query) - DuckDuckGo web search
• analyze_url(url) - Extract and analyze webpage content
• arxiv_search(query) - Academic paper search with full PDF text

Financial & Economic Data

• get_stock_overview(symbol) - Real-time stock, crypto, and market data
• analyze_canadian_economy() - Economic indicators and analysis

Media & Content

• analyze_youtube_url(url) - Video transcription and analysis

Location & Weather

• get_weather(location) - Weather forecasts by city or coordinates
• get_tide_info(location) - Canadian coastal tide times

Toronto Data

• get_toronto_crime(neighbourhood, crime_type) - Crime statistics by area
• list_toronto_neighbourhoods() - Available neighborhoods

Development

Local Setup

# Core dependencies
uv sync

# With Streamlit UI
uv sync --extra local

# With development tools
uv sync --extra dev

Testing Commands

# Streamlit UI
uv run streamlit run app.py

# MCP server for Claude Desktop
uv run python mcp_server.py

# HTTP mode testing
uv run python -m src.server http 8000

Environment Variables

• YOUTUBE_MAX_TOKENS=24000 - Transcript processing limit
• WEBSHARE_PROXIES="ip:port:user:pass,..." - Proxies for YouTube cloud deployment
• MCP_RETRY_MAX_ATTEMPTS=3 - Auto-retry failed tool calls
• MCP_RETRY_TYPE_COERCION=true - Auto-fix type mismatches

Tool Development

Tools in src/tools/ modules use @mcp.tool(description="...") decorator with automatic schema generation from Python type hints.

Deployment

Cloud (FastMCP)

1. Push to main branch
2. FastMCP Cloud auto-deploys from GitHub
3. Get auth token from dashboard
4. Configure clients with deployment URL

Local Development

1. Clone repository
2. uv sync --extra local
3. uv run streamlit run app.py

Documentation

• DEPLOYMENT.md - Deployment guide
• CLAUDE.md - Development notes
• FastMCP Cloud
• MCP Protocol