[](https://oPipeline[](https://opensource.org/licenses/MIT)

This repository demonstrates how to combine Model Context Protocol (MCP) and Agent-to-Agent (A2A) machinery with an LLM backend (such as Ollama) to perform JSON-Schema generation. It showcases a minimal, end-to-end example for practical implementation.

The core idea is to offload schema generation from an LLM, leveraging MCP's structured communication and A2A's routing capabilities. This allows for dynamic and automated creation of JSON schemas from raw JSON data.

• Python 3.8+
• Ollama: (https://ollama.com/) running locally on http://localhost:11434. Make sure you have downloaded at least one model (e.g., gemma3:12b).
• pip: For installing Python packages.

.
├── agents
│   ├── schema_agent.py       # A2A agent that calls the MCP tool
│   └── schema_tool.py        # FastMCP server exposing the 
`analyze_schema` tool
│
├── data
│   └── out.json              # Sample JSON input for testing
│
├── main.py                   # (Optional) Launcher to start all 
components at once
├── orchestrator
│   └── orchestrator.py       # A2A “router” that forwards .json messages 
to the schema agent
│
├── client_interactive.py     # Simple command-line client to talk to the 
orchestrator
└── requirements.txt          # Python dependencies

• MCP Server (agents/schema_tool.py):

  • Role: Hosts a single MCP tool named analyze_schema.
  • Functionality: Instantiates a FastMCP server, registers the analyze_schema tool (which generates schemas from JSON strings), and builds a FastAPI application exposing JSON-RPC endpoints under /a2a.
  • Endpoint: JSON-RPC available at http://localhost:8001/a2a. Also supports REST endpoints under /tools/analyze_schema (configurable).

• Schema Agent (agents/schema_agent.py):

  • Role: An A2A agent that handles JSON file analysis requests.
  • Functionality: Connects to the MCP server's JSON-RPC endpoint, scans incoming text for .json substrings, reads referenced files from data/, calls the MCP tool via call_mcp_tool, and returns schemas as chat responses.
  • Endpoint: A2A messages served at http://localhost:8002/a2a.

• Orchestrator (orchestrator/orchestrator.py):

  • Role: Routes messages to either the schema agent or the LLM.
  • Functionality: Inherits from an Ollama-backed A2A server. Checks for .json in incoming text, forwards to the schema agent's A2A endpoint, otherwise uses the LLM directly.
  • Endpoint: A2A messages served at http://localhost:8000/a2a.

• Sample Data (data/out.json):

  • Role: Provides a simple JSON file for demonstration.
  • Content: A minimal JSON object (e.g., with numeric, array, nested object fields).

• Launcher Script (main.py):

  • Role: Convenience script to start all three services.
  • Functionality: Spawns the MCP server, schema agent, and orchestrator as background processes.

1. Ensure Ollama is running: Launch it with a model (e.g., ollama serve gemma3:12b --api-url http://localhost:11434).
2. Install Python dependencies:

   pip install -r requirements.txt

3. Verify file existence: Make sure you have:
   • agents/schema_tool.py
   • agents/schema_agent.py
   • orchestrator/orchestrator.py
   • data/out.json
   • main.py (optional)
   • client_interactive.py
   • requirements.txt

1. Start Ollama: (See Installation & Setup)

2. Start the MCP Server:

   uvicorn agents.schema_tool:app --host 0.0.0.0 --port 8001

   (JSON-RPC endpoint: http://localhost:8001/a2a)

3. Start the Schema Agent:

   python3 agents/schema_agent.py

   (A2A server on port 8002)

4. Start the Orchestrator:

   python3 orchestrator/orchestrator.py

   (A2A server on port 8000)

5. Use the Interactive Client:

   python3 client_interactive.py

• Switch between JSON-RPC and REST: Modify agents/schema_tool.py to expose only REST endpoints and update agents/schema_agent.py accordingly.
• Modify the LLM prompt: Edit the Ollama system prompt in agents/schema_tool.py to adjust schema formatting (e.g., draft-04 vs. draft-07).
• Add additional MCP tools: Register more functions with @mcp.tool() in agents/schema_tool.py.

• Ollama Version: Ensure compatibility between your Ollama version and the code.
• Port Conflicts: If ports 8000, 8001, or 8002 are already in use, change the corresponding arguments in the startup commands.
• JSON Schema Draft: The LLM's ability to produce valid JSON schemas may vary. You might need to refine the prompt.