Privacy and capability boundaries for agent access
The public agent services are designed to explain the site, not to take custody of documents.
What the A2A and MCP services can access
They receive a text question and return links, summaries, suggested next steps, and documented limits from a small public page index. The services are appropriate for questions such as “Does this converter support OCR?” or “Where is the Markdown table guide?”
What they cannot access
- They do not accept a PDF, Markdown file, image, URL to fetch, or document bytes.
- They do not perform server-side PDF conversion, OCR, visual layout analysis, or remote export.
- They do not retain a conversation or offer a task status endpoint.
Why the boundary matters
The ordinary conversion tools run in a visitor's browser. A remote agent-upload endpoint would be a different product with different data handling, deletion, abuse prevention, authentication, and consent requirements. The current services intentionally do not imply that such an endpoint exists.
Choose the safe workflow
- Use A2A or MCP to find the relevant public instructions.
- Have the user open the local browser tool and select their own file.
- For scans, choose an OCR-capable workflow before expecting Markdown output.
- Review the exported Markdown or PDF before sharing it.
For the technical request format, see the A2A calling guide. For browser file handling, see what runs locally in the PDF-to-Markdown converter.