Needs minor changes

The docs translation looks solid and the MCP fix is well-tested. A few issues worth addressing:

Code issues

1. filter parameter accepted but never used (src/memos/api/mcp_serve.py:273-299)

The filter param is added to the signature and normalized to None, but it's never passed to self.mos_core.search():

if not filter:
    filter = None
result = self.mos_core.search(query, user_id, cube_ids)  # filter not forwarded

If this is intentional (just to prevent MCP clients from erroring when they send filter: {}), add a comment making that explicit. If it should actually filter results, wire it through.

2. filter shadows the built-in (src/memos/api/mcp_serve.py:276)

Using filter as a parameter name shadows Python's built-in. Consider search_filter or filters instead. Minor, but it can confuse linters and developers.

3. if not filter is overly broad (src/memos/api/mcp_serve.py:297)

not filter is truthy for {}, None, [], 0, "", etc. If the intent is specifically "empty dict or None → treat as no filter," be explicit:

if filter is None or filter == {}:
    filter = None

This avoids accidentally swallowing a filter value of 0 or False if the type annotation ever loosens.

Documentation nits

4. Typo in memos_neo.md — the embedding dimension is stated as 3027:

维度为 text-embedding-3-large（dim-size 3027）

OpenAI's text-embedding-3-large default dimension is 3072. If this is a project-specific truncation, clarify; otherwise fix the number.

5. memos_mcp.md JSON has a trailing comma (line ~96 of the new file):

"--transport", "stdio"
],
//   "cwd": "/path/to/your/MemOS pip user is optional",

The commented-out cwd line with // isn't valid JSON. This will confuse users who copy-paste. Either remove it or use a note outside the code block.

Test feedback

Tests are clean and cover the key cases well. One suggestion: add a test with a non-empty filter dict (e.g., filter={"tag": "work"}) to document the current behavior (that it's silently ignored). This prevents future regressions if someone later wires filter through.

PR hygiene

The PR description says "Fixes #1691" but #1691 appears to be the PR itself, not a separate issue. Double-check the linked issue number.

Overall the translation quality is good — class names and code stay in English as promised, and the prose reads naturally.