[mangeshraut712]

What changed

An unfiltered POST /v1/search from Notion-Version 2025-09-03 on lists each database as its data source, the object GET /v1/data_sources/{id} already returns, and lists no database object. 2022-06-28 still lists databases. A database object filter is unchanged, because that filter under the newer versions is not in the measurement. Closes #722.

Why this is what the real API does

Measured against api.notion.com on 2026-10-14, recorded in #722: POST /v1/search with {} returns page and database under 2022-06-28, and page and data_source under 2025-09-03 and 2026-03-11. Each data source is that database's own, in the shape GET /v1/data_sources/{id} answers less its request_id. A {"filter":{"property":"object","value":"data_source"}} search already listed data sources; an unfiltered search on those versions now uses that same object.

Verification

• Tests — uv sync --all-extras --locked, then uv pip install --no-deps "llama-index-readers-hubspot<0.6", then uv run --no-sync pytest -rs: 10326 passed, 1 skipped, 5 warnings in 113.88s. The skip is tests/test_mcp.py:154: Docker not available. test_notion_unfiltered_search_lists_a_database_in_the_versions_shape failed on 2025-09-03 and 2026-03-11 before this change and passes after it; 2022-06-28 passed before and after.

• Optional surfaces that ran rather than skipped — official SDK, llama-index, MCP (except the Docker case), mirage, and fsspec extras are installed, and those tests ran.

• Lint — ruff check . printed All checks passed!. ruff format --check . printed 159 files already formatted. ruff check --preview --select PLW1514 . printed All checks passed!. python scripts/prose_check.py upstream/main exited 0. backlot diff --source notion printed 56 divergence(s) · 56 acknowledged · 0 new, so the Notion baseline is unchanged.

• A test fails without this change

• A new endpoint serving corpus content is ACL-scoped, proved for both an admin and a scoped token — no new endpoint; search still returns only documents the caller can see

• Comments state what was measured, not the history of the fix

  

[nuriyoo]

Thank you for this; the unfiltered search now lists each database's data source from 2025-09-03 on, and the data source it lists is the one GET /v1/data_sources/{id} answers. I measured the searches again on 2026-10-17 (api.notion.com, an integration's workspace of three databases and ten pages) and every search this branch changes lists the objects real lists. The suggestions are a comment that also records a query and the shape, and the new test as rows of the table above it.

Two neighbours the body can name rather than call unmeasured. Under 2025-09-03 and 2026-03-11 real refuses a database filter with a 400 validation_error (body.filter.value should be `"page"` or `"data_source"`, instead was `"database"`.), which is #375. And every search from 2025-09-03 on ends with "type":"page_or_data_source", which is #374.

On the suggestion tree the whole suite passes (10326, 2 skipped: Docker and the HubSpot reader), and 12 of the 15 mutations generated from the diff fail a test. The three left: swapping the two operands of render_data_source, which gives the same value; dropping want in (None, "data_source"), which only changes the database filter under 2025-09-03, a request real refuses; and query.strip(), on a line this PR does not touch.

[nuriyoo]

Thank you, @mangeshraut712. The unfiltered search now lists each database's data source from 2025-09-03 on, and the object it lists is the one GET /v1/data_sources/{id} answers.

I finished it with a maintainer commit (b8a0a2f) from the review: the comment on search records the query case and the shape with the 2026-10-17 measurement, and the search rows are one table that checks every listed object against its own GET. The Notion tests pass on the branch updated with main, and CI is green.