CasePro Documents

CasePro Documents

The connector finds the user's documents and gives you a link to read each one. Reading respects the user's own document permissions.

First, make sure the connector is available. If the search_documents / get_document tools aren't present, the CasePro connector isn't enabled for this chat — ask the user to enable the CasePro connector and sign in. Use only these tools to reach documents; do not try to open the CasePro website in a browser.

Tools

• search_documents — search the user's documents.
  • query: keywords. filter: all (default), files, folders, content, tags. Use content to search text inside documents.
  • Optional: folderId, mimeTypes (e.g. application/pdf), page, pageSize.
  • Returns each match with a file_id.
• get_document — fetch a document by file_id. Returns the document's metadata and a secure, time-limited download link.
• discuss_document — give it a file_id OR a query (picks the best match) and it returns the document to discuss.

How to actually read a document

get_document returns the document. Small documents come back ready to read. Larger ones come with a secure download link — to read those you download the file and open it, which works best in Claude Cowork or Claude Code (anywhere you can run code and save files).

1. Call get_document (or discuss_document) to get the document (or its download link).
2. If the document is returned directly, just read it.
3. If you got a download link, download the file into your workspace and open it — for example, in Cowork or Claude Code, use the code tool to fetch the link to a local file (curl/requests) and then read it with a PDF/text library.

Reading large documents (bigger than ~30 MB)

In Claude Cowork a single file can't be loaded straight into the conversation if it's over ~30 MB, but the code environment can still handle it — download it to a local file and process it there, without pulling the whole thing into the conversation:

• Download the file to disk first (don't try to read 50 MB into context at once).
• Then read it with code: open the PDF and pull page ranges or specific sections (e.g. the police-report pages, the billing summary), or split it into smaller parts and read the part you need. This keeps each step small while still letting you read the whole document across steps.
• Search within the downloaded file for the keywords that matter (party names, "citation", "diagnosis", dollar amounts) and read around the hits.

In Claude Code there's no size limit — download and read the whole file directly.

If you genuinely can't open a document in the current environment (for example, a plain chat without file access), say so simply and suggest the user open it in Claude Cowork, where you can download and read it — then keep going with whatever you can do from the structured case data.

Notes

• Download links are time-limited — if one stops working, just call get_document again to get a fresh one.
• Always tell the user which document you read, by its name.
• If the user doesn't have access to a document, say so simply and move on — don't keep retrying.