fix packages/markitdown/src/markitdown/_uri_utils.py

• from urllib.parse import urlparse, unquote_to_bytes

• from urllib.parse import urlparse, unquote, unquote_to_bytes

• path = os.path.abspath(url2pathname(parsed.path))

• decoded_path = unquote(parsed.path)
• path = os.path.abspath(url2pathname(decoded_path))

Before the change, file_uri_to_path() passed parsed.path directly into url2pathname().

That usually works for simple ASCII file paths, but it breaks when the file URI contains percent-encoded non-ASCII characters such as Korean filenames. In our case, a URI like:

file:///D:/.../%EC%A0%9C20...hwpx

was not being decoded into the original Unicode path before conversion to a Windows filesystem path. As a result, the generated path became invalid and MCP failed to open the file.

The fix was to explicitly call unquote(parsed.path) first, so the percent-encoded URI path is restored to its real Unicode form before url2pathname() converts it into a local OS path.

In short:

• before: URI path stayed percent-encoded too long
• after: URI path is decoded first, then converted to a Windows path

That makes file: URIs with Unicode filenames work correctly in the MCP flow.

Open packages/markitdown/src/markitdown/_uri_utils.py and inspect file_uri_to_path(), starting with how parsed.path is passed to url2pathname(). Verify the change with a file URI containing a percent-encoded Unicode filename, and confirm the resulting Windows path lets the MCP flow open the file correctly.