Research-Stack/4-Infrastructure/shim/gemma4_mcp.py

#!/usr/bin/env python3
"""
gemma4_mcp.py — Local Gemma4-12B MCP server.

Calls the local llama-server endpoint at http://127.0.0.1:8081/v1.
No API key required. Returns reasoning + content.

Usage:
  python3 gemma4_mcp.py   # starts MCP server on stdio
"""

import json
import sys

try:
    import httpx
except ImportError:
    httpx = None

GEMMA_URL = "http://127.0.0.1:8081/v1/chat/completions"
GEMMA_MODEL = "gemma4-12b"


def call_gemma(question: str, system: str = "", max_tokens: int = 2000,
               temperature: float = 0.3, enable_thinking: bool = True) -> dict:
    """Call local Gemma4-12B model."""
    if httpx is None:
        return {"error": "httpx not installed"}

    messages = []
    if system:
        messages.append({"role": "system", "content": system})
    messages.append({"role": "user", "content": question})

    payload = {
        "model": GEMMA_MODEL,
        "messages": messages,
        "max_tokens": max_tokens,
        "temperature": temperature,
    }
    if not enable_thinking:
        payload["chat_template_kwargs"] = {"enable_thinking": False}

    try:
        with httpx.Client(timeout=120) as client:
            resp = client.post(
                GEMMA_URL,
                json=payload,
                headers={
                    "Content-Type": "application/json",
                    "Authorization": "Bearer none",
                },
            )
            resp.raise_for_status()
            data = resp.json()

        msg = data["choices"][0]["message"]
        content = msg.get("content", "")
        reasoning = msg.get("reasoning_content", "")
        usage = data.get("usage", {})
        timings = data.get("timings", {})

        return {
            "content": content,
            "reasoning": reasoning,
            "tokens_in": usage.get("prompt_tokens", 0),
            "tokens_out": usage.get("completion_tokens", 0),
            "prompt_speed": round(timings.get("prompt_per_second", 0), 1),
            "generation_speed": round(timings.get("predicted_per_second", 0), 1),
        }
    except Exception as e:
        return {"error": str(e)}


def mcp_server():
    for line in sys.stdin:
        try:
            req = json.loads(line.strip())
        except json.JSONDecodeError:
            continue

        method = req.get("method", "")
        req_id = req.get("id")
        params = req.get("params", {})

        if method == "initialize":
            resp = {
                "jsonrpc": "2.0", "id": req_id,
                "result": {
                    "protocolVersion": "2024-11-05",
                    "capabilities": {"tools": {}},
                    "serverInfo": {"name": "gemma4", "version": "0.1.0"},
                },
            }
        elif method == "tools/list":
            resp = {
                "jsonrpc": "2.0", "id": req_id,
                "result": {
                    "tools": [{
                        "name": "gemma4",
                        "description": (
                            "Ask a question to the local Gemma4-12B model. "
                            "Free, fast (~40 tok/s), good at math and code. "
                            "Use this for quick questions, sanity checks, "
                            "and mathematical reasoning."
                        ),
                        "inputSchema": {
                            "type": "object",
                            "properties": {
                                "question": {
                                    "type": "string",
                                    "description": "The question to ask",
                                },
                                "system": {
                                    "type": "string",
                                    "description": "System prompt (optional)",
                                },
                                "max_tokens": {
                                    "type": "number",
                                    "description": "Max response tokens (default: 2000)",
                                },
                                "enable_thinking": {
                                    "type": "boolean",
                                    "description": "Enable reasoning mode (default: true)",
                                },
                            },
                            "required": ["question"],
                        },
                    }],
                },
            }
        elif method == "tools/call":
            tool_name = params.get("name", "")
            arguments = params.get("arguments", {})

            if tool_name == "gemma4":
                result = call_gemma(
                    question=arguments.get("question", ""),
                    system=arguments.get("system", ""),
                    max_tokens=arguments.get("max_tokens", 2000),
                    enable_thinking=arguments.get("enable_thinking", True),
                )
            else:
                result = {"error": f"Unknown tool: {tool_name}"}

            resp = {
                "jsonrpc": "2.0", "id": req_id,
                "result": {
                    "content": [{"type": "text", "text": json.dumps(result, indent=2)}],
                },
            }
        else:
            resp = {
                "jsonrpc": "2.0", "id": req_id,
                "error": {"code": -32601, "message": f"Method not found: {method}"},
            }

        print(json.dumps(resp), flush=True)


if __name__ == "__main__":
    mcp_server()