mcp-memory-service v8.19.0: retrieve_memory and search_by_tag broken with 'tags' and 'float' errors

4 answers

• Best answerAgentArmor teamTeam3d ago

  Status: fixed (releases v8.19.1 and v8.20.1). The two v8.19.0 regressions were addressed by the maintainer: search_by_tag was fixed in v8.19.1 (commit 10b8c57), and a follow-up retrieve_memory serialization bug introduced by that patch was addressed in v8.20.1 (commit d1609a7). The thread reports no user confirmation of v8.20.1, only the release announcement.

  Root causes:

  • retrieve_memory: MemoryService.retrieve_memories() was calling storage.retrieve() with tags and memory_type parameters that the base MemoryStorage.retrieve(query, n_results) interface does not accept. v8.19.1 removed those unsupported arguments and moved tag/memory-type filtering into the service layer after retrieval.
  • search_by_tag: the handler assumed created_at is always an ISO-format string, but the hybrid (sqlite + Cloudflare) backend returns a float Unix timestamp, so .replace('Z', '+00:00') crashed with AttributeError: 'float' object has no attribute 'replace'. v8.19.1 added type checks: floats go through datetime.fromtimestamp(), strings through datetime.fromisoformat().

  Important caveat: v8.19.1's retrieve_memory fix introduced a second error, 'MemoryQueryResult' object has no attribute 'content', because the response formatter expected a Memory object but received a MemoryQueryResult. The maintainer fixed it in v8.20.1 by extracting .memory before formatting and adding similarity_score to the response.

  To update:

  bashCopy to clipboard

  pip install --upgrade mcp-memory-service
  # or
  uv pip install --upgrade mcp-memory-service
  

  Then restart the MCP server (/mcp in Claude Code, or restart Claude Desktop).

  Workaround: on the reporter's v8.19.0 setup (Windows, Python 3.12, hybrid backend, Claude Desktop), recall_memory, store_memory, and check_database_health still worked, so recall_memory can be used while broken.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What a GitHub user tried on GitHub (Didn't work)

  Upgraded from v8.18.2 to v8.19.0 and called retrieve_memory(query="upgrade protocol venv312", n_results=3) and search_by_tag(tags=["mcp-memory-service", "upgrade"]). Both failed with the reported errors; recall_memory worked as a workaround.

  Source: https://github.com/doobidoo/mcp-memory-service/issues/211 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @VibeCodeChef tried on GitHub (Partly worked)

  After upgrading to v8.19.1, verified that search_by_tag was fixed, but retrieve_memory now failed with a new error: 'MemoryQueryResult' object has no attribute 'content'.

  Source: https://github.com/doobidoo/mcp-memory-service/issues/211#issuecomment-3504729935 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @doobidoo tried on GitHub (Outcome unclear)

  Maintainer released v8.20.1 (commit d1609a7) on PyPI with a fix that extracts the Memory object from MemoryQueryResult before formatting and adds similarity_score. No user in the thread confirmed the fix after this release.

  Source: https://github.com/doobidoo/mcp-memory-service/issues/211#issuecomment-3506717237 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport