Make doc preprocessing options configurable in MCP server

[kanlac]

Summary

Add --use_doc_orientation_classify and --use_doc_unwarping CLI arguments (and corresponding PADDLEOCR_MCP_USE_DOC_ORIENTATION_CLASSIFY / PADDLEOCR_MCP_USE_DOC_UNWARPING environment variables) to the MCP server.

Previously these options were hardcoded to False with no way for users to override them. The default remains False to preserve backward compatibility.

Problem

When processing scanned PDF documents via the MCP server in service mode (aistudio, self_hosted), layout detection can miss entire text blocks — lines are silently dropped from the output with no error or warning. Enabling useDocUnwarping and useDocOrientationClassify in the API request resolves the issue.

Since these options are hardcoded to False and not exposed as configuration, users have no way to work around this without patching the source code.

Related: #17164

Usage

# Via environment variables
PADDLEOCR_MCP_USE_DOC_UNWARPING=true paddleocr_mcp ...

# Via CLI arguments
paddleocr_mcp --use_doc_unwarping --use_doc_orientation_classify ...

Changes

• mcp_server/paddleocr_mcp/__main__.py: Add two new CLI arguments with env var fallbacks
• mcp_server/paddleocr_mcp/pipelines.py: Replace hardcoded False with configurable instance variables

[CLAassistant]

All committers have signed the CLA.

[paddle-bot]

Thanks for your contribution!

[Bobholamovic]

Could you also update the documentation accordingly? Also, would it make more sense to let the agent decide whether to apply doc preprocessing, rather than fixing these settings at server init?