MCP Agent Gateway

git clone https://github.com/wuwo1979/agent.git
cd agent
pip install -r requirements.txt

# Run demo (recommended first)
python demo.py

# Run benchmarks
python main.py --benchmark

# Start gateway
python main.py --host 0.0.0.0 --port 9090

User/App -> MCP Gateway (JSON-RPC 2.0) -> 14 Built-in Tools
                     |
             Supervisor-Worker <- LangGraph
                     |
             Performance: Cache | Parallel | Adapter

$ python demo.py

================================================================
  MCP Gateway + Multi-Agent: Codebase Health Check
================================================================

  Scenario: Developer needs to analyze a Python project
  - 14 Python files across 4 modules
  - Without MCP: 5+ custom scripts, 3 tool integrations
  - With MCP:    1 unified pipeline, 3 lines of config

  [1] Creating test codebase (14 Python files)...
    OK Created 14 files, 402 total lines

  [2] Registering MCP tool providers...
    OK 3 providers, 11 tools ready

  [3] Testing individual MCP tools...
    OK list_dir found 4 items (2.9ms)
    OK read_file 683 chars (2.3ms)
    OK sysinfo windows (245.9ms)
    OK run_command 'hello' (182.1ms)

  [5] Performance analysis...
    Cache hit rate: 66.7%

================================================================
  Results Summary
================================================================
  Tools registered   | 14  | 3 providers
  Cache hit rate     | 66.7% | repeated reads use cache
  Files analyzed     | 14  | 402 lines, 5 with TODOs

agent/
├── mcp_gateway/          # MCP Gateway Layer
│   ├── protocol.py       # Protocol core + ToolRegistry
│   ├── transport.py      # HTTP/SSE/STDIO transport
│   ├── server.py         # Production gateway entry
│   ├── security.py       # Auth + rate limiting
│   └── tools/            # 14 built-in tools
│       ├── filesystem.py # Filesystem (5 tools)
│       ├── terminal.py   # Terminal (3 tools)
│       └── database.py   # Database (5 tools)
├── agent_scheduler/      # Agent Scheduling Layer
│   ├── graph.py          # LangGraph workflow
│   ├── supervisor.py     # Supervisor-Worker pattern
│   ├── state.py          # State + snapshot management
│   ├── retry.py          # Retry + circuit breaker
│   └── agents/           # 3 Worker agents
│       ├── planner.py    # Task decomposition
│       ├── executor.py   # Tool execution
│       └── validator.py  # Result verification
├── performance/          # Performance Layer
│   ├── cache.py          # Incremental context cache
│   ├── parallel.py       # Parallel scheduler + DAG
│   └── adapter.py        # Multi-model adapter
├── core/                 # Infrastructure
│   ├── types.py          # Data types
│   ├── interfaces.py     # Abstract interfaces
│   ├── exceptions.py     # Error hierarchy
│   └── observability.py  # Metrics + health check
├── docker/               # Docker deployment
│   ├── Dockerfile
│   └── docker-compose.yml
├── tests/                # Tests
│   ├── test_mcp.py       # 10 MCP tests
│   ├── test_agent.py     # 12 Agent tests
│   └── benchmark.py      # 5 benchmarks
├── docs/                 # Documentation
│   ├── MCP Protocol.md
│   ├── A2A Protocol.md
│   ├── LangGraph Guide.md
│   ├── Performance.md
│   └── Interview Prep.md
├── demo.py               # Demo script
├── main.py               # Entry point
└── requirements.txt