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

[llA1ll]

• 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.

[Jah-yee]

Good day,

I noticed this issue and wanted to share a fix. The problem is that file_uri_to_path() passes parsed.path directly to url2pathname() without decoding percent-encoded characters first.

For URIs like , the percent-encoded Unicode path stays encoded when passed to url2pathname(), resulting in an invalid filesystem path.

Here is the simple fix:

In packages/markitdown/src/markitdown/_uri_utils.py:

1. Add import:

2. Decode the path before url2pathname():

This decodes the percent-encoded URI path back to its original Unicode form before url2pathname() converts it to a local OS path.

Thank you for your work on this project. I hope this small fix is helpful. Please let me know if there's anything to adjust.

Warmly, RoomWithOutRoof