A2A.AIAgent

# A2A AI Agent — Multi-Agent Research Assistant An automated research pipeline powered by Claude AI. Give it a question, and a team of coordinated AI agents searches the web, analyzes findings, and produces a full markdown report — all observable in real time. <img width="1551" height="742" alt="image" src="https://github.com/user-attachments/assets/31e806fb-4c7d-4fb1-8e78-1221efdcf4d8" /> <img width="1554" height="749" alt="image" src="https://github.com/user-attachments/assets/76804689-f6e3-4fcc-bb4c-fe0fceaa4e28" /> <img width="1525" height="750" alt="image" src="https://github.com/user-attachments/assets/5b1e1c10-7f35-404c-869f-dc347be85ad5" /> --- ## How It Works ``` User Query ↓ [FastAPI Backend] ←→ [React Frontend] ↓ [Orchestrator Agent] (A2A Protocol) ├─→ [Search Agent] → MCP Web Search → Raw Results ├─→ [Analyst Agent] → MCP File System → Key Insights └─→ [Writer Agent] → MCP File System → Final Report ↓ reports/ (markdown output) ``` Four specialist agents communicate over the **Agent-to-Agent (A2A) protocol** via HTTP. Tool calls (web search, file I/O) go through **MCP servers**. The frontend streams every event live via WebSocket. --- ## Tech Stack | Layer | Technology | |-------|-----------| | AI Agents | Claude (Anthropic) via `anthropic` SDK | | Agent Protocol | A2A (custom HTTP-based) | | Tool Protocol | MCP (Model Context Protocol) | | Web Search | Tavily API | | Backend | FastAPI + WebSockets | | Frontend | React 19, Vite, Tailwind CSS, Framer Motion | --- ## Prerequisites - Python 3.10+ - Node.js 18+ - [Anthropic API key](https://console.anthropic.com/) - [Tavily API key](https://tavily.com/) --- ## Setup ### 1. Clone & configure environment ```bash git clone https://github.com/Harshprajapat777/A2A.AIAgent.git cd A2A.AIAgent cp .env.example .env ``` Edit `.env` and fill in your keys: ```env ANTHROPIC_API_KEY=your_anthropic_key TAVILY_API_KEY=your_tavily_key ``` ### 2. Install backend dependencies ```bash pip install -r requirements.txt ``` ### 3. Install frontend dependencies ```bash cd frontend npm install ``` --- ## Running **Backend** (from project root): ```bash uvicorn api.main:app --reload --port 8080 ``` **Frontend** (in a second terminal): ```bash cd frontend npm run dev ``` Open [http://localhost:3000](http://localhost:3000). --- ## Environment Variables | Variable | Required | Default | Description | |----------|----------|---------|-------------| | `ANTHROPIC_API_KEY` | Yes | — | Powers all four Claude agents | | `TAVILY_API_KEY` | Yes | — | Web search capability | | `MODEL_NAME` | No | `claude-sonnet-4-6` | Claude model to use | | `API_PORT` | No | `8080` | FastAPI backend port | | `MCP_WEB_SEARCH_PORT` | No | `8001` | Web search MCP server port | | `MCP_FILE_SYSTEM_PORT` | No | `8002` | File system MCP server port | | `SEARCH_AGENT_PORT` | No | `8010` | Search agent A2A port | | `ANALYST_AGENT_PORT` | No | `8011` | Analyst agent A2A port | | `WRITER_AGENT_PORT` | No | `8012` | Writer agent A2A port | | `ORCHESTRATOR_PORT` | No | `8013` | Orchestrator agent A2A port | --- ## Project Structure ``` A2A/ ├── agents/ # Four AI agents (orchestrator, search, analyst, writer) ├── a2a/ # A2A protocol layer (registry, task manager, message types) ├── api/ # FastAPI backend (routes, websocket streaming, report endpoints) ├── config/ # Settings loader (reads .env, validates required keys) ├── mcp_servers/ # MCP tool servers (web search, file system) ├── frontend/ # React app (dashboard, report viewer, live pipeline view) ├── reports/ # Generated markdown research reports ├── requirements.txt └── .env.example ``` --- ## API Endpoints | Method | Path | Description | |--------|------|-------------| | `POST` | `/research` | Start a new research job | | `GET` | `/reports` | List all saved reports | | `GET` | `/reports/{id}` | Retrieve a specific report | | `WS` | `/ws/stream/{job_id}` | Stream live agent events | | `GET` | `/health` | Health check | --- ## Frontend Pages - **Home** — Search box + report history - **Dashboard** — Live pipeline view: agent diagram, A2A call log, MCP call log, activity feed - **Report** — Rendered markdown report with export button --- ## License MIT