Want help regarding mcp server how to connect wtih llm agent and render the resukt to frontend

[vedantv3]

I am sending my prompt prompt should go to backend then to mcp server how to connect it with llm so that I can get the result and render back same on frontend

I'm building a backend workflow using FastMCP on Windows to connect a local AI model via CLI. My goal is to send prompts from a frontend (via JSON-RPC) to MCP, which then invokes a model subprocess and returns the response.
I’ve tried integrating Cursor CLI (cursor-agent), but it’s not supported on Windows (Git Bash returns Unsupported operating system: MINGW64_NT-10.0-26100, and PowerShell says 'cursor-agent' is not recognized).
So I switched to Ollama, which runs fine on Windows. I installed it and can run ollama run mistral manually.
In my MCP server, I use this subprocess call:
process = subprocess.Popen(
["ollama", "run", "mistral"],
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
text=True,
shell=True
)
stdout, stderr = process.communicate(input=prompt, timeout=30)
But when I send a prompt from the frontend, I get a blank response:
{"jsonrpc":"2.0","result":"","id":...}

✅ MCP logs show:

• Prompt received
• Subprocess launched
• stdout is empty
• stderr is empty

❓ What I need help with:

• Why is ollama run mistral returning nothing when called via subprocess?
• Is there a better way to invoke Ollama from Python on Windows?
• How can I confirm that MCP is correctly capturing the model output?
  Any help or working examples would be appreciated!