fix(mcp): harden list_cubes member arrays

[Bartok9]

What failure does this repair?

MCP list_cubes raised AttributeError: 'str' object has no attribute 'get' when a cube's measures/dimensions/time_dimensions was non-list or held non-dict members. Repro: call the list_cubes tool against a manifest whose cube has "measures": ["revenue"] — name extraction called .get("name") on the string and the tool failed.

Summary

list_cubes now ignores invalid cube entries and extracts names only from valid dictionary members.

Verification

pytest tests/unit/test_mcp_server.py::test_list_cubes_skips_non_dict_members -q
# 1 passed

Duplicate check

Searched open/closed PRs; #2615 hardens related cube listing but not the member-array normalization in this path. No overlap.

[coderabbitai]

Walkthrough

list_cubes now handles malformed cube and member structures defensively, while a unit test verifies filtering and name extraction behavior.

Changes

Cube enumeration robustness

Layer / File(s)	Summary
Normalize cube members and validate malformed input core/wren/src/wren/mcp_server.py, core/wren/tests/unit/test_mcp_server.py	list_cubes ignores invalid cube entries and extracts names only from valid dictionary members; the unit test covers malformed measures and mixed dimensions.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• Canner/WrenAI#2615: Related hardening of cube listing and tests for malformed cube members.

Poem

I hop through cubes, both neat and strange,
And skip the shapes that drift or change.
Valid names sparkle in my tray,
While malformed bits bounce away.
Tests thump softly—hip-hop, hooray!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title is concise and accurately describes the main change: hardening list_cubes member array handling.
Description check	✅ Passed	The description covers the required sections with a clear failure, summary, test verification, and duplicate check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[goldmedal]

Reviewing this together with #2615 — the MCP half of one change.

#2615 and #2617 harden the same cube listing behind two entry points — the wren cube list CLI and the list_cubes MCP tool. They should be one PR, because right now each covers a case the other misses and one of them guards states its data source cannot produce.

The two data sources differ, and that decides which guards are live

MCP list_cubes reads context.load_cubes, which already normalises. Both _load_cubes_v1 and _load_cubes_v2 do if isinstance(data, dict) before appending, and both always return a list. Measured against hand-edited cube YAML (a file that is a bare scalar, a file that is a list, and a valid cube with measures: nope):

load_cubes -> [{'name': 'orders', 'measures': 'nope', 'dimensions': [{'name': 'id'}, 'x'], ...}]
  container is list?        True     # `isinstance(cubes, list)` always holds
  any non-dict cube?        False    # `isinstance(cube, dict)` always holds
  any non-list 'measures'?  True     # member arrays ARE unvalidated

So in #2617 the two cube-level guards are dead and only _member_names is real. The test reaches the dead branches only by monkeypatching wren.context.load_cubes to return ["bad", ...] — a value the real loader cannot emit.

The CLI reads mdl.json through json.loads (_load_manifest_dict), which validates only that the top level is an object. Everything below it is genuinely untrusted, so #2615's cube-level guard is live. But the container is not guarded, and that is a silent-wrong-output bug. Measured on main + #2615:

cubes: "abc"        -> exit 0, output ''                      # neither listed nor "No cubes defined."
cubes: []           -> exit 0, output 'No cubes defined.'
cubes: ["bad", {…}] -> exit 0, lists orders                   # fixed by this PR

A scalar cubes prints a blank screen that is indistinguishable from neither-of-the-above. #2617 guards exactly this container case; #2615 does not.

What to do

Fold the two into one PR that:

• guards the container and the member arrays in both entry points, so the CLI and the MCP tool agree
• drops the cube-level guard and its monkeypatched test from the MCP side, where load_cubes already guarantees list[dict]
• states in the description which side of the wren-core deserialization boundary each callsite sits on — the CLI path is upstream (json.loads), the MCP path is behind a loader that normalises

Better still, follow 9bdae39 (#2604): if member arrays should be trustworthy, normalise them once in load_cubes and report malformed ones from validate_project, rather than re-checking at every consumer.

Rebase

Both branches are 17 commits behind main and their CI ran against a base that predates 9bdae39. They merge cleanly, but a textually clean merge is not a semantically correct one — please rebase and re-run.

[Bartok9]

Thanks @goldmedal — agreed on the data-source boundary analysis.

Plan:

1. Close this PR and fix(cube): harden list against non-dict cube/measure rows #2615 in favor of one combined follow-up that covers both entry points.
2. Prefer the fix(context): filter non-mapping entries in _load_views_v1 #2604 / 9bdae39c pattern: normalise member arrays in load_cubes (and guard the CLI mdl.json container/list[dict] path where json.loads is upstream of normalisation), with validate_project reporting malformed member arrays rather than re-checking at every consumer.
3. Drop dead cube-level guards + monkeypatched load_cubes returns on the MCP side; keep live member-array handling and the CLI container guard.
4. Rebase onto current main (post-9bdae39c) and re-run CI before opening the replacement.

Will open the consolidated PR shortly and link both of these from it.