Markitdown MCP Plugin for VSCode

[norsemangrey]

I'm attempting to use the Markitdown MCP server plugin in VSCode so that I can hoock it up to a custom agent and do batch conversion of documents. However, there is very little documentation of the MCP version in VSCode and I am not sure how to configure it properly. For instance it does not seem that it is able to convert "non-selectable text" documents or part of documents. There does not seem to be an option to configure OCR for the pluging, but I though perhaps that was a "native" feature in this environment as it is already "hooked" up to the Github Copilot in a sense. Maybe someone else is using this and can help clarify?

[Tongyi (TongyiDai)]

The VS Code MCP connection and OCR are two separate layers here. The stock markitdown-mcp server currently exposes only:

convert_to_markdown(uri)

and constructs MarkItDown with only the plugin-enable flag. It does not expose llm_client, llm_model, or Azure Document Intelligence/Content Understanding settings as MCP tool parameters. Copilot being the MCP client does not implicitly add OCR to the server.

That means:

• Text PDFs should work with the ordinary converter.
• Scanned PDFs/non-selectable text need a separate OCR path.
• Setting MARKITDOWN_ENABLE_PLUGINS=true alone is insufficient for the markitdown-ocr plugin, because that plugin also needs an LLM client and vision-capable model; without them it silently falls back to the normal converter.

For the current release, the simplest batch workflow is to install markitdown-ocr in a virtual environment and run the CLI outside MCP:

python -m venv .venv
source .venv/bin/activate
pip install 'markitdown[all]' markitdown-ocr openai
markitdown scanned.pdf --use-plugins --llm-client openai --llm-model gpt-4o -o scanned.md

Then let the VS Code agent consume the generated Markdown. If you specifically need OCR through MCP, a small custom wrapper/fork is currently required so the server can construct MarkItDown(enable_plugins=True, llm_client=..., llm_model=...). The built-in server cannot be configured that way today.

Relevant implementation/docs:

• https://github.com/microsoft/markitdown/tree/main/packages/markitdown-mcp
• https://github.com/microsoft/markitdown/tree/main/packages/markitdown-ocr