diff --git a/main.py b/main.py
index 3394e02..5327258 100644
--- a/main.py
+++ b/main.py
@@ -40,7 +40,7 @@ def gather_handler_kwargs(handler_name: str) -> dict:
 
 
 async def process_dump(
-    mappings: dict[str, str], handlers, max_concurrent: int
+    mappings: dict[str, str], handlers
 ):
     """
     Stream-download the bzip2-compressed XML dump and feed to SAX.
@@ -52,7 +52,7 @@ async def process_dump(
     )
     decomp = bz2.BZ2Decompressor()
     sax_parser = xml.sax.make_parser()
-    dump_handler = WikiDumpHandler(mappings, handlers, max_concurrent)
+    dump_handler = WikiDumpHandler(mappings, handlers)
     sax_parser.setContentHandler(dump_handler)
 
     async with aiohttp.ClientSession() as session:
@@ -75,9 +75,18 @@ async def main():
         logger.error("Error: set ENV HANDLER (e.g. 'filesystem' or 'filesystem,sftp')")
         sys.exit(1)
 
+    # 2. Read concurrency setting
+    try:
+        max_conc = int(os.getenv("MAX_CONCURRENT", "0"))
+    except ValueError:
+        raise ValueError("MAX_CONCURRENT must be an integer")
+
+    if max_conc < 0:
+        raise ValueError("MAX_CONCURRENT must be >= 0")
+        
     handlers = []
     
-    # 2. Load each handler
+    # 3. Load each handler
     for handler_name in handler_names:
         handler_name = handler_name.strip()
         if not handler_name:
@@ -102,20 +111,14 @@ async def main():
 
         # Build kwargs from ENV
         handler_kwargs = gather_handler_kwargs(handler_name)
+        
+        # Add max_concurrent to kwargs
+        handler_kwargs["max_concurrent"] = max_conc
 
         # Instantiate
         handler = HandlerCls(**handler_kwargs)
         handlers.append(handler)
 
-    # 3. read concurrency setting
-    try:
-        max_conc = int(os.getenv("MAX_CONCURRENT", "0"))
-    except ValueError:
-        raise ValueError("MAX_CONCURRENT must be an integer")
-
-    if max_conc < 0:
-        raise ValueError("MAX_CONCURRENT must be >= 0")
-
     # 4. Fetch mappings
     logger.info("Fetching mappings from SQL dump…")
     mappings = await fetch_mappings()
@@ -123,7 +126,7 @@ async def main():
 
     # 5. Stream & split the XML dump
     logger.info("Processing XML dump…")
-    await process_dump(mappings, handlers, max_conc)
+    await process_dump(mappings, handlers)  # Pass 0 as max_concurrent since handlers handle it
 
     # 6. Finish up
     await asyncio.gather(*[handler.close() for handler in handlers])
diff --git a/output_handlers/base_handler.py b/output_handlers/base_handler.py
index bd8f20b..706b085 100644
--- a/output_handlers/base_handler.py
+++ b/output_handlers/base_handler.py
@@ -1,6 +1,7 @@
 """Reference handler for output handlers."""
 from abc import ABC, abstractmethod
 import logging
+import asyncio
 
 
 
@@ -14,15 +15,20 @@ class BaseHandler(ABC):
     _successful_writes = 0
     _failed_writes = 0
 
-    def __init__(self, fail_on_error: bool = True, **kwargs):
+    def __init__(self, fail_on_error: bool = True, max_concurrent=0, **kwargs):
         """
         Initializes the BaseHandler with optional parameters.
 
         Args:
             fail_on_error (bool): If True, the handler will raise an exception on error. Defaults to True.
+            max_concurrent: Maximum number of concurrent write operations.
+                            0 means unlimited concurrency.
             **kwargs: Additional keyword arguments for specific handler implementations.
         """
         self.fail_on_error = fail_on_error
+        self.semaphore = None
+        if max_concurrent > 0:
+            self.semaphore = asyncio.Semaphore(max_concurrent)
 
 
     @abstractmethod
@@ -47,7 +53,11 @@ class BaseHandler(ABC):
             entry (dict): The entry to write (will be JSON-encoded).
             uid (str): The unique identifier for the entry. The default id provided by wikivoyage is recommended. 
         """
-        success = await self._write_entry(entry, uid)
+        if self.semaphore:
+            async with self.semaphore:
+                success = await self._write_entry(entry, uid)
+        else:
+            success = await self._write_entry(entry, uid)
         if success:
             self.logger.debug(f"Successfully wrote entry with UID {uid}")
             self._successful_writes += 1
diff --git a/transformers/wiki_dump_handler.py b/transformers/wiki_dump_handler.py
index 48a1c7e..c566f44 100644
--- a/transformers/wiki_dump_handler.py
+++ b/transformers/wiki_dump_handler.py
@@ -12,14 +12,11 @@ class WikiDumpHandler(xml.sax.ContentHandler):
     and write via the user‐supplied handler(s).
     """
 
-    def __init__(self, mappings, handlers, max_concurrent):
+    def __init__(self, mappings, handlers):
         super().__init__()
         self.mappings = mappings
         # Support a single handler or a list of handlers
         self.handlers = handlers
-        self.sem = (
-            asyncio.Semaphore(max_concurrent) if max_concurrent > 0 else None
-        )
         self.tasks: list[asyncio.Task] = []
 
         self.currentTag: str | None = None
@@ -55,10 +52,7 @@ class WikiDumpHandler(xml.sax.ContentHandler):
                 title = self.currentTitle
                 logger.debug(f"scheduled {wd_id} for handling")
                 # schedule processing
-                if self.sem:
-                    task = asyncio.create_task(self._bounded_process(text, wd_id, title))
-                else:
-                    task = asyncio.create_task(self._process(text, wd_id, title))
+                task = asyncio.create_task(self._process(text, wd_id, title))
                 self.tasks.append(task)
             else:
                 logger.debug(f"page {pid} without wikidata id, skipping...")
@@ -104,8 +98,3 @@ class WikiDumpHandler(xml.sax.ContentHandler):
         await asyncio.gather(*[
             handler.write_entry(entry, uid) for handler in self.handlers
         ])
-
-    async def _bounded_process(self, text: str, uid: str, title: str):
-        # Only run N at once
-        async with self.sem:
-            await self._process(text, uid, title)