Agent Swarm Platform

A distributed multi-agent system where specialized agents collaborate to achieve complex objectives. Built on Microsoft Foundry with MCP (Model Context Protocol) server integration.

Key Principles

• Distributed Intelligence: Tasks are divided among specialized agents, enabling parallel processing and efficiency
• Emergent Behavior: Interactions among agents create sophisticated solutions beyond individual capabilities
• Adaptive Response: The swarm dynamically adjusts to changing conditions, ensuring resilience
• Collaborative Learning: Agents share insights and improve collectively via historical execution plan search

Architecture

User → Swarm Controller → Planner Agent → Worker Agents → Aggregator Agent → Result
            ↕                    ↕               ↕               ↕
       MCP Server          AI Search        MCP Tools        SQLite DB
            ↕                    ↕               ↕               ↕
      Azure Storage        Vector DB      Foundry SDK    App Insights

Component	Port	Description
swarm_controller	8000	Central orchestrator – task distribution, lifecycle management
planner_agent	8001	Decomposes objectives into executable task plans
swarm_registry	8002	Catalog of agents, MCP servers, tools, and skills
aggregator_agent	8003	Synthesizes task results into final output
swarm_mcp_server_foundry	8004	MCP server exposing swarm operations
swarm_dashboard	8080	Web UI for monitoring swarm status

Communication Layer

• Azure Storage Queues – asynchronous messaging between agents and controller
• Azure Blob Storage – binary file sharing between agents
• FastAPI HTTP – synchronous service-to-service communication

Quick Start

Prerequisites

• Python 3.12+
• Azure subscription (OpenAI, AI Search, Storage, Container Apps)
• Azure Developer CLI

Local Development

# Install dependencies
pip install -r requirements.txt

# Copy and configure environment
cp .env.example .env

# Run all services (each in a separate terminal)
cd src/swarm_controller && uvicorn app:app --port 8000
cd src/planner_agent && uvicorn app:app --port 8001
cd src/swarm_registry && uvicorn app:app --port 8002
cd src/aggregator_agent && uvicorn app:app --port 8003
cd src/swarm_mcp_server_foundry && python server.py
cd src/swarm_dashboard && uvicorn app:app --port 8080

Deploy to Azure

azd auth login
azd up

# Deploy individual services
./azd-hooks/deploy.sh swarm-controller <env-name>
./azd-hooks/deploy.sh planner-agent <env-name>
./azd-hooks/deploy.sh swarm-registry <env-name>
./azd-hooks/deploy.sh aggregator-agent <env-name>
./azd-hooks/deploy.sh swarm-mcp-server-foundry <env-name>
./azd-hooks/deploy.sh swarm-dashboard <env-name>

Project Structure

├── agents.md                          # Agent capabilities documentation
├── azure.yaml                         # Azure Developer CLI configuration
├── docs/                              # Documentation
│   ├── architecture.md
│   └── getting-started.md
├── infra/                             # Infrastructure as Code (Bicep)
│   ├── main.bicep                     # Main template
│   ├── ai/                            # OpenAI + AI Search
│   ├── core/host/                     # Container Apps + Registry
│   ├── core/monitor/                  # Application Insights
│   ├── core/security/                 # Key Vault + RBAC
│   └── core/storage/                  # Storage Account
├── skills/                            # Skill definitions (markdown)
│   ├── research-and-summarize.md
│   ├── code-generation.md
│   ├── data-analysis.md
│   ├── document-synthesis.md
│   └── multi-step-workflow.md
└── src/
    ├── swarm_controller/              # Central orchestrator
    ├── planner_agent/                 # Plan decomposition
    ├── aggregator_agent/              # Result aggregation
    ├── swarm_registry/                # Agent/tool registry
    ├── swarm_mcp_server_foundry/      # MCP server interface
    └── swarm_dashboard/               # Web dashboard

License

See LICENSE for details.