Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

fix(anthropic): treat data URL scheme and media type case-insensitively

lei_leiab806eba4dad
diff --git a/browser_use/llm/anthropic/serializer.py b/browser_use/llm/anthropic/serializer.py
index 834b3efda..f52d065f3 100644
--- a/browser_use/llm/anthropic/serializer.py
+++ b/browser_use/llm/anthropic/serializer.py
@@ -29,20 +29,30 @@ class AnthropicMessageSerializer:
@staticmethod
def _is_base64_image(url: str) -> bool:
- """Check if the URL is a base64 encoded image."""
- return url.startswith('data:image/')
+ """Check if the URL is a base64 encoded image.
+
+ URI scheme names are case-insensitive (RFC 3986). MIME type/subtype
+ names are case-insensitive (RFC 2045). Compare the lowered scheme and
+ media type so DATA:image/PNG;base64,... is still treated as a data URL.
+ """
+ scheme, sep, remainder = url.partition(':')
+ if not sep or scheme.lower() != 'data':
+ return False
+ media_type = remainder.split(',', 1)[0].split(';', 1)[0].lower()
+ return media_type.startswith('image/')
@staticmethod
def _parse_base64_url(url: str) -> tuple[SupportedImageMediaType, str]:
"""Parse a base64 data URL to extract media type and data."""
- # Format: data:image/jpeg;base64,<data>
- if not url.startswith('data:'):
+ # Format: data:image/jpeg;base64,<data> (scheme and media type case-insensitive)
+ scheme, sep, remainder = url.partition(':')
+ if not sep or scheme.lower() != 'data':
raise ValueError(f'Invalid base64 URL: {url}')
- header, data = url.split(',', 1)
- media_type = header.split(';')[0].replace('data:', '')
+ header, data = remainder.split(',', 1)
+ media_type = header.split(';', 1)[0].lower()
- # Ensure it's a supported media type
+ # Ensure it's a supported media type (canonical lowercase for Anthropic)
supported_types = ['image/jpeg', 'image/png', 'image/gif', 'image/webp']
if media_type not in supported_types:
# Default to jpeg if not recognized
diff --git a/tests/ci/models/test_anthropic_data_url_case.py b/tests/ci/models/test_anthropic_data_url_case.py
new file mode 100644
index 000000000..e5b8208a4
--- /dev/null
+++ b/tests/ci/models/test_anthropic_data_url_case.py
@@ -0,0 +1,58 @@
+import pytest
+
+from browser_use.llm.anthropic.serializer import AnthropicMessageSerializer
+from browser_use.llm.messages import ContentPartImageParam, ImageURL, UserMessage
+
+PNG_DATA = 'iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=='
+JPEG_DATA = '/9j/4AAQSkZJRgABAQAAAQABAAD/2wAAA=='
+GIF_DATA = 'R0lGODlhAQABAIAAAAAAAP///ywAAAAAAQABAAACAUwAOw=='
+WEBP_DATA = 'UklGRiIAAABXRUJQVlA4IBYAAAAwAQCdASoBAAEADsD+JaQAA3AAAAAA'
+HTTPS_URL = 'https://example.com/cat.png'
+
+
+def serialize_image(url: str):
+ message = UserMessage(content=[ContentPartImageParam(image_url=ImageURL(url=url))])
+ serialized = AnthropicMessageSerializer.serialize(message)
+ assert isinstance(serialized['content'], list)
+ return serialized['content'][0]
+
+
+ 'url',
+ [
+ f'data:image/png;base64,{PNG_DATA}',
+ f'data:image/PNG;base64,{PNG_DATA}',
+ f'DATA:image/png;base64,{PNG_DATA}',
+ f'Data:Image/Png;base64,{PNG_DATA}',
+ ],
+)
+def test_data_url_case_variants_remain_png_base64_images(url):
+ block = serialize_image(url)
+ assert block['source'] == {
+ 'type': 'base64',
+ 'media_type': 'image/png',
+ 'data': PNG_DATA,
+ }
+
+
+ 'media_type,data',
+ [
+ ('image/jpeg', JPEG_DATA),
+ ('image/png', PNG_DATA),
+ ('image/gif', GIF_DATA),
+ ('image/webp', WEBP_DATA),
+ ],
+)
+def test_supported_media_types_keep_canonical_lowercase(media_type, data):
+ block = serialize_image(f'data:{media_type.upper()};base64,{data}')
+ assert block['source'] == {
+ 'type': 'base64',
+ 'media_type': media_type,
+ 'data': data,
+ }
+
+
+def test_remote_https_image_stays_a_url_source():
+ block = serialize_image(HTTPS_URL)
+ assert block['source'] == {'type': 'url', 'url': HTTPS_URL}