MCP: `find_bookmarks` returns "Unknown error" for specific `collection_id` values

[axelquack]

Summary: Filtering find_bookmarks by a specific collection_ids value intermittently returns "Unknown error" with no further detail. The direct REST API (GET /raindrops/{id}) works reliably for the same collection.

• Expected: Consistent results when filtering by collection.
• Actual: Intermittent failures depending on the collection. Observed consistently with larger collections.
• Workaround: Use GET /raindrops/{collection_id} via REST API instead.
• Related: https://github.com/axelquack/raindrop-mcp-scripts

[exentrich]

Can you show me your exact request and what response you receive?

[axelquack]

Thanks for following up. Here's the exact request we used:

 {
   "jsonrpc": "2.0",
   "id": 1,
   "method": "tools/call",
   "params": {
     "name": "find_bookmarks",
     "arguments": {
       "collection_ids": [22803193],
       "limit": 50
     }
   }
 }

The response at the time was:

{ "result": { "isError": true, "content": [{ "type": "text", "text": "Unknown error" }] } }

I've since tried to reproduce the issue and can no longer do so. Testing find_bookmarks with collection_ids on several collections now - including one with over 1,700 bookmarks - all return results correctly. The INBOX collection itself (same ID, now with 11 items) also works fine.

At the time of the original report, INBOX had 554 bookmarks and I was actively running bulk move operations against it - moving items out in batches of 100 via update_bookmarks. My best guess is that the error was caused by a race condition or write-lock contention: querying a collection while it was being heavily modified concurrently may have put the backend in an inconsistent state, resulting in the vague "Unknown error" response.

If that's the case, the real issue is the error message itself - "Unknown error" gives no actionable signal. Even a generic "collection temporarily unavailable, retry" would be more useful. Happy to close this if you can confirm that concurrent read/write on the same collection is a known edge case.