Problem

The MCP server's convert_pdf tool (python/opendataloader-pdf-mcp/) still exposes the legacy --format values markdown-with-html and markdown-with-images in its format parameter. These were split out of --format in the core CLI by #508 (PDFDLOSP-6) in favor of explicit flags:

• HTML-in-Markdown → --markdown-with-html
• image extraction → --image-output off|embedded|external

The core CLI keeps the old values for one major release with a deprecation warning, for backward compatibility. But MCP is a newer surface (introduced in #351) with no installed client base depending on the old tokens, so there's no reason to carry them — they just confuse agents about the correct interface.

There's also a latent bug: for format=\"markdown\" with no image_output, images default to external and get written into the temp directory the MCP server uses internally, which is then discarded — so the returned Markdown references images that no longer exist.

Proposed change (MCP server only)

• Remove markdown-with-html / markdown-with-images from the accepted format values. Valid values: json, text, html, markdown.
• Add an explicit markdown_with_html: bool parameter (maps to --markdown-with-html).
• For markdown/html output, default image_output to embedded when the caller doesn't specify it, so images survive the temp-dir round-trip. Explicit values still win.
• Update README and tests.

Out of scope

Core java/, node/, and python/opendataloader-pdf keep the deprecated tokens per the one-major-release compatibility policy established in #508. The node / python legacy code paths that still emit the old tokens only do so from already-deprecated run()-style helpers, so they aren't user-facing; they'll be cleaned up when those helpers are removed.

Acceptance criteria

• convert_pdf(format=\"markdown-with-html\") raises ValueError
• convert_pdf(format=\"markdown-with-images\") raises ValueError
• markdown_with_html=True forwards --markdown-with-html to the JAR
• format=\"markdown\" defaults to image_output=\"embedded\"; explicit value overrides
• format=\"json\" does not set a default image_output
• README reflects the new interface

References

• fix(cli): separate markdown modifiers from --format values (PDFDLOSP-6) #508 — fix(cli): separate markdown modifiers from --format values (PDFDLOSP-6) (core split; MCP not touched)
• feat: add MCP server for AI agent integration #351 — feat: add MCP server for AI agent integration (original MCP surface)