multi-agent-mcp / examples /claude_code_usage.py
Cduplar's picture
Initial public release: Multi-Agent MCP Delegation Server
8b02e7c
Raw
History Blame
5.87 kB
"""
Example: How Claude Code should use the Delegation MCP Server
This demonstrates the CORRECT MCP code execution pattern:
βœ… Claude Code writes code that imports and uses the MCP server
❌ NOT a chat UI that directly calls MCP tools
When you ask Claude Code to delegate a task, it should write code similar to this.
"""
import asyncio
from pathlib import Path
from delegation_mcp.server import DelegationMCPServer
from delegation_mcp.config import DelegationConfig
async def example_delegation():
"""Example: Claude Code delegating a security audit to Gemini."""
print("πŸš€ Initializing Delegation MCP Server...")
# Initialize the server (Claude Code does this)
server = DelegationMCPServer(
config_path=Path("config/delegation_rules.yaml"),
enable_security=True,
enable_persistence=True,
)
print("βœ… Server initialized\n")
# Example 1: Delegate security audit to Gemini
print("πŸ“ Example 1: Security Audit")
print("Query: 'Audit auth.py for SQL injection vulnerabilities'")
print("Expected: Routes to Gemini (best for security analysis)\n")
# This is what Claude Code would do when you ask it to delegate:
result = await server.engine.process(
"Audit the authentication code for SQL injection vulnerabilities"
)
print(f"Orchestrator: {result.orchestrator}")
print(f"Delegated to: {result.delegated_to}")
print(f"Success: {result.success}")
print(f"Duration: {result.duration:.2f}s")
print(f"Output preview: {result.output[:200]}...\n")
print("-" * 60 + "\n")
# Example 2: Delegate refactoring to Claude
print("πŸ“ Example 2: Code Refactoring")
print("Query: 'Refactor database connection to use connection pooling'")
print("Expected: Routes to Claude (best for architecture)\n")
result = await server.engine.process(
"Refactor the database connection code to use connection pooling"
)
print(f"Orchestrator: {result.orchestrator}")
print(f"Delegated to: {result.delegated_to}")
print(f"Success: {result.success}")
print(f"Duration: {result.duration:.2f}s")
print(f"Output preview: {result.output[:200]}...\n")
print("-" * 60 + "\n")
# Example 3: Check agent statistics
print("πŸ“Š Example 3: Get Delegation Statistics")
if server.persistence:
stats = server.persistence.get_statistics()
print(f"Total tasks: {stats.get('total_tasks', 0)}")
print(f"Success rate: {stats.get('success_rate', 0):.1%}")
print(f"Average duration: {stats.get('avg_duration', 0):.2f}s")
print(f"Agent usage: {stats.get('agent_usage', {})}\n")
print("βœ… Examples complete!")
print("\n" + "=" * 60)
print("πŸ’‘ Key Insight:")
print("=" * 60)
print("""
This is the CORRECT MCP pattern:
- Claude Code writes and runs this Python code
- The code imports and uses the delegation MCP server
- Tasks are routed to specialized agents automatically
- Results come back through the code
This is WRONG:
- A Gradio chat UI that directly calls MCP tools
- That violates the code execution pattern
- The UI should only monitor activity, not execute it
""")
async def example_mcp_protocol_usage():
"""
Example: How an MCP client (like Claude Code) uses the server via MCP protocol.
Note: This is pseudocode showing what happens under the hood when Claude Code
uses the MCP server through the stdio protocol.
"""
print("\n" + "=" * 60)
print("πŸ“‘ MCP Protocol Usage (Pseudocode)")
print("=" * 60 + "\n")
print("""
When Claude Code is configured to use this MCP server:
1. Configuration (~/.config/claude/mcp.json):
{
"mcpServers": {
"delegation": {
"command": "delegation-mcp"
}
}
}
2. User asks Claude Code: "Audit my code for security issues"
3. Claude Code recognizes it should use the delegation server
4. Claude Code calls the MCP tool via stdio:
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "delegate_task",
"arguments": {
"query": "Audit auth.py for SQL injection",
"orchestrator": "claude"
}
}
}
5. Delegation MCP server receives the call, routes to Gemini
6. Response comes back:
{
"orchestrator": "claude",
"delegated_to": "gemini",
"success": true,
"output": "Found 3 SQL injection vulnerabilities...",
"duration": 2.5
}
7. Claude Code presents the result to the user
βœ… User works with Claude Code, but gets Gemini's security expertise!
""")
if __name__ == "__main__":
print("""
╔═══════════════════════════════════════════════════════════════╗
β•‘ Delegation MCP Server - Correct Usage Pattern Examples β•‘
β•‘ β•‘
β•‘ This demonstrates how Claude Code (or other MCP clients) β•‘
β•‘ should use the delegation server following Anthropic's β•‘
β•‘ code execution pattern. β•‘
β•šβ•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•
""")
# Run async examples
asyncio.run(example_delegation())
# Show MCP protocol usage
asyncio.run(example_mcp_protocol_usage())
print("\n" + "=" * 60)
print("🎯 Ready for Production!")
print("=" * 60)
print("""
To use in production:
1. Install: pip install -e .
2. Configure Claude Code to use delegation-mcp
3. Chat with Claude Code naturally
4. Watch tasks get routed to the best agent!
Optional: Run `delegation-monitor` to visualize activity for demos.
""")