Fixed workspace volume mount

This commit is contained in:
Morpheus Sandmann
2026-07-15 19:00:59 +01:00
parent 09b17505e8
commit 61898f3359
2 changed files with 30 additions and 25 deletions
+1 -1
View File
@@ -1,7 +1,7 @@
FROM python:3.11-slim
# Install system dependencies including git for version control operations
RUN apt-get update && apt-get install -y git && rm -rf /var/lib/apt/lists/* && pip install fastmcp mcp requests playwright markitdown && playwright install chromium --with-deps
RUN apt-get update && apt-get install -y git && rm -rf /var/lib/apt/lists/* && pip install fastmcp mcp requests playwright markitdown[all] && playwright install chromium --with-deps
RUN groupadd -g 1000 user && \
useradd -m -u 1000 -g 1000 -s /bin/bash user
+29 -24
View File
@@ -12,6 +12,7 @@ Provides 4 tools:
import os
import sys
import re
import json
import logging
import asyncio
import subprocess
@@ -26,12 +27,29 @@ from starlette.middleware.cors import CORSMiddleware
from starlette.responses import JSONResponse
# ---------------------------------------------------------------------------
# Logging
# Logging (JSON Format)
# ---------------------------------------------------------------------------
class JSONLogFormatter(logging.Formatter):
"""Custom JSON formatter for professional structured logging."""
def format(self, record):
log_record = {
"timestamp": self.formatTime(record, self.datefmt),
"level": record.levelname,
"logger": record.name,
"message": record.getMessage(),
"file": f"{record.filename}:{record.lineno}"
}
if record.exc_info:
log_record["exception"] = self.formatException(record.exc_info)
return json.dumps(log_record)
json_handler = logging.StreamHandler(sys.stdout)
json_handler.setFormatter(JSONLogFormatter())
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(name)s - %(message)s",
handlers=[logging.StreamHandler(sys.stdout)],
handlers=[json_handler],
force=True # Forces Uvicorn/FastAPI to also use this JSON handler
)
logger = logging.getLogger("turnstone-mcp")
@@ -47,9 +65,10 @@ PLAYWRIGHT_TEXT_TIMEOUT = int(os.environ.get("PLAYWRIGHT_TEXT_TIMEOUT", "15000")
EXTRACT_MAX_FILE_SIZE = int(os.environ.get("EXTRACT_MAX_FILE_SIZE", "10485760")) # 10 MB
WORKSPACE_ROOT = os.path.abspath(os.environ.get("WORKSPACE_ROOT", os.getcwd()))
# Defaults to /workspace to match Docker volumes
WORKSPACE_ROOT = os.path.abspath(os.environ.get("WORKSPACE_ROOT", "/workspace"))
logger.info("Workspace Root: %s", WORKSPACE_ROOT)
logger.info(f"Workspace Root configured to: {WORKSPACE_ROOT}")
# ---------------------------------------------------------------------------
# FastMCP app
@@ -76,8 +95,6 @@ class PlaywrightBrowserManager:
logger.info("Starting headless Playwright session…")
# CRITICAL: We must store the context_manager as a class property.
# If it's garbage collected, it forces the Playwright driver to close.
self.context_manager = async_playwright()
self.playwright = await self.context_manager.start()
@@ -120,16 +137,16 @@ class ToolObserverMiddleware(Middleware):
async def on_call_tool(self, context: MiddlewareContext, call_next):
tool_name = getattr(context.message, "name", "unknown")
arguments = getattr(context.message, "arguments", {})
logger.info("📥 [LLM REQUEST] Invoking tool: '%s'", tool_name)
logger.info(" ↳ Arguments: %s", arguments)
logger.info(f"LLM REQUEST | Tool: '{tool_name}' | Args: {arguments}")
try:
result = await call_next(context)
content = getattr(result, "content", result)
logger.info("📤 [SERVER RESPONSE] Tool '%s' completed.", tool_name)
logger.info(" ↳ Content: %s\n", str(content)[:500])
preview = str(content)[:250].replace('\n', ' ') # Clean up newlines for JSON reading
logger.info(f"SERVER RESPONSE | Tool: '{tool_name}' completed | Preview: {preview}")
return result
except Exception as e:
logger.error("❌ [SERVER ERROR] Tool '%s' crashed: %s", tool_name, e)
logger.error(f"SERVER ERROR | Tool '{tool_name}' crashed: {e}")
raise
mcp.add_middleware(ToolObserverMiddleware())
@@ -176,9 +193,6 @@ def execute_python_code(code: str) -> str:
if len(code) > 50_000:
return "Error: code exceeds maximum length (50,000 chars)"
# --- Memory limit via injected script ---
# We inject the resource limit directly into the subprocess script.
# Applying it in the main process would incorrectly limit the MCP server!
memory_limit_code = f"""
try:
import resource
@@ -189,7 +203,6 @@ except Exception:
"""
full_code = memory_limit_code.strip() + "\n\n" + code
# --- Timeout via subprocess ---
try:
with tempfile.NamedTemporaryFile(
mode="w", suffix=".py", delete=False
@@ -213,7 +226,6 @@ except Exception:
except OSError:
pass
# --- Output handling ---
stdout = proc.stdout[:PYTHON_MAX_OUTPUT]
stderr = proc.stderr[:PYTHON_MAX_OUTPUT]
@@ -242,7 +254,6 @@ async def playwright_navigate(url: str) -> str:
Navigation result including page title, or error message.
"""
try:
# --- Validate URL ---
parsed = urlparse(url)
if not parsed.scheme or not parsed.netloc:
if re.match(r"^[a-zA-Z0-9]", url):
@@ -279,8 +290,6 @@ async def playwright_get_page_text() -> str:
if not page:
return "Error: no active browser session. Run playwright_navigate first."
# CRITICAL FIX: page.evaluate does not support a `timeout` argument.
# We must wrap the call in asyncio.wait_for instead.
text_content = await asyncio.wait_for(
page.evaluate("() => document.body.innerText"),
timeout=PLAYWRIGHT_TEXT_TIMEOUT / 1000.0,
@@ -324,20 +333,16 @@ def extract_to_markdown(relative_path: str) -> str:
if not os.path.isfile(target):
return f"Error: file not found at {relative_path}"
# --- Size check ---
file_size = os.path.getsize(target)
if file_size > EXTRACT_MAX_FILE_SIZE:
return f"Error: file size ({file_size} bytes) exceeds limit ({EXTRACT_MAX_FILE_SIZE} bytes)"
# --- Extension check ---
ext = os.path.splitext(target)[1].lower()
if ext and ext not in SUPPORTED_EXTENSIONS:
return f"Error: unsupported file type '{ext}'. Supported: {', '.join(sorted(SUPPORTED_EXTENSIONS))}"
# --- Convert ---
try:
from markitdown import MarkItDown
mid = MarkItDown()
result = mid.convert(target)
return result.text_content