使用说明
ProBound AI Orchestrator (MCP Edition)
A high-concurrency AI agent backend utilizing the Model Context Protocol (MCP) to standardize and scale agent-to-tool communication. This project was developed as a research-focused extension to demonstrate advanced infrastructure for resilient, high-throughput AI services. 🚀 Architectural Advantages
-
Model Context Protocol (MCP) Integration Instead of traditional REST-based tool calls, this orchestrator acts as an MCP Host. It communicates with a dedicated MCP Tool Server via JSON-RPC over Standard I/O (stdio). This decouples the "thinking" (API logic) from the "doing" (Tool execution), allowing for LLM-agnostic tool discovery.
-
Zero Network Overhead By utilizing a persistent pipe between the Host and the Tool Server, we eliminated the 15% failure rate observed in standard REST architectures caused by external API throttling and TCP handshake overhead.
-
Non-Blocking Orchestration Leveraging Python's asyncio and the mcp SDK, the orchestrator manages 5,000 simultaneous tool calls without blocking the FastAPI event loop, ensuring the system remains responsive even when integrations are slow.
🛠 Setup & Execution Prerequisites Docker & Docker Compose
Python 3.11+ (for local testing)
Launching the System Bash
Build and start the orchestrator and load tester
docker-compose up --build -d Running the Stress Test Access the Locust UI at http://localhost:8089.
Configure the test:
Users: 1000
Spawn Rate: 100
Host: http://mcp-orchestrator:8000
Click Start Swarming and monitor the real-time performance graphs.
📁 Repository Structure /app: FastAPI Orchestrator (MCP Host) logic.
/mcp_service: The Tool Provider (MCP Server) hosting the integration tools.
locust.py: Performance benchmarking suite.
Dockerfile & docker-compose.yml: Standardized deployment and scaling environment.
