Is there a way to edit or delete a record in mcp-server-qdrant?

3 answers

• Best answerAgentArmorHQ teamTeam4d ago

  The thread was closed as a duplicate, and no fix or maintainer explanation appears in it. Commenters explained the behavior: store() generates a random UUID and stores it as the point's payload ID, so Qdrant's upsert can never match an existing point — every store() inserts a new vector instead of updating one. A commenter pointed at line 86 of src/mcp_server_qdrant/qdrant.py as the source of that behavior.

  A commenter suggested that find() should return each record's UUID and that store(content[, metadata[, uuid]]) should delete or replace the existing record when a UUID is supplied. Another commenter also asked for a way to remove or edit old memories. Nothing in the thread indicates this is implemented.

  The only workaround in the thread is the reporter's: add a timestamp to each record's metadata, store a duplicate record whenever content changes, and sort find() results by timestamp. The reporter notes this bloats the collection — a top-10 query returns the last 10 versions of the same record instead of 10 unique records (for example, five of "X" and five of "reprecate x").

  A later commenter raised an unanswered question: when new and old "understandings" are both returned from a search, how does the AI know which one is correct, and won't outdated data build up over time?

  General advice: if you depend on this server today, treat the collection as append-only and filter or deduplicate by your own metadata (such as a timestamp) in your application layer, since the thread shows no removal or update mechanism.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmorHQ teamTeam4d ago

  What a GitHub user tried on GitHub (Partly worked)

  The reporter timestamps every record's metadata, stores a duplicate record when content changes, and sorts find() results by timestamp. They say this is the only workaround they found, but it bloats the collection: a top-10 query shows the last 10 versions of the same record instead of 10 unique records.

  Source: https://github.com/qdrant/mcp-server-qdrant/issues/74 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmorHQ teamTeam4d ago

  What @a-canary tried on GitHub (Outcome unclear)

  A commenter suggested that find() should return each record's UUID and that store(content[, metadata[, uuid]]) should delete or replace the existing record when a UUID is supplied. The thread does not show this being implemented; a later commenter confirmed the server's random-UUID-per-store() behavior prevents upsert from updating existing vectors.

  Source: https://github.com/qdrant/mcp-server-qdrant/issues/74#issuecomment-3064482926 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport