luke-dump

Luke Dump Skill

Exhaustive pagination of Notion data sources via the bundled luke-notion MCP server. Uses the Notion REST API directly with cursor-based pagination — no 100-row cap.

When to use

• User asks for a complete enumeration ("ALL of X", "EVERY Y").
• User asks for a count where completeness matters ("how many tasks total in Backlog").
• User asks for an export or full-board audit.
• Another skill (typically luke-tasks) needs to enumerate a domain.

When NOT to use

• Targeted single-task lookup → use hosted notion-fetch.
• Keyword discovery → use hosted notion-search.
• Known-small result sets (<100 rows) → hosted notion-query-database-view is fine.

Canonical data source IDs

DB	Data source URI
Tasks	collection://b0d00fd8-eebb-434d-84c1-a652260fbe79
Meetings	collection://db2e5ca6-92b7-4e39-ab23-f4c496d2636c
Initiatives	collection://28a0a1b7-d639-4e34-898f-e19415823dec

Invocation patterns

Count-only

For "how many X in Y status" queries — paginate fully but return just the number.

mcp__luke-notion__dump-data-source({
  data_source_id: "collection://b0d00fd8-eebb-434d-84c1-a652260fbe79",
  filter: { "property": "Status", "select": { "equals": "Backlog" } },
  count_only: true
})

Response: {"count": 237}.

Full dump to file

For large enumerations — write to a file to keep the tool response small, then read + process the file.

mcp__luke-notion__dump-data-source({
  data_source_id: "collection://b0d00fd8-eebb-434d-84c1-a652260fbe79",
  filter: { "property": "Status", "select": { "equals": "Backlog" } },
  output_path: "/tmp/luke-dump-backlog.json"
})

Response: {"count": 237, "path": "/tmp/luke-dump-backlog.json"}. Then Read the file or use Bash with jq to slice/aggregate.

Inline (only for small result sets)

For enumerations expected to return <200 rows. Returns the full array in the tool response.

mcp__luke-notion__dump-data-source({
  data_source_id: "collection://28a0a1b7-d639-4e34-898f-e19415823dec"
})

Response: {"count": N, "rows": [...]}.

Partial returns

If pagination hits a persistent 429 or 5xx mid-dump (after one retry), the server returns partial: true in the response with whatever rows it successfully fetched. ALWAYS check for partial before reporting counts as authoritative.

{"count": 412, "partial": true, "path": "/tmp/x.json"}  // got 412 of an unknown total

In that case, surface the partial state to the user:

"Got 412 rows before hitting a rate limit. Count is incomplete — re-run to get the rest."

Filter shape

The filter JSON is passed through verbatim to POST /v1/data_sources/{id}/query. Follow the Notion data_sources endpoint schema (may differ slightly from the legacy databases endpoint — consult https://developers.notion.com/reference/query-a-data-source when composing non-trivial filters).

Common filter patterns:

• Status equals: {"property": "Status", "select": {"equals": "Backlog"}}
• Initiative relation is-empty: {"property": "Initiative", "relation": {"is_empty": true}}
• Compound AND: {"and": [filterA, filterB]}
• Compound OR: {"or": [filterA, filterB]}

Errors and what they mean

• "NOTION_TOKEN is not set" — the user hasn't provided the token. Direct them to run /plugin install luke-notion@luke-plugins to re-prompt, or export NOTION_TOKEN=ntn_... in shell.
• "integration lacks access to this data source" — token is set but the integration isn't connected to the target DB. Tell the user to open the DB in Notion → ••• → Connections → Add luke-notion.
• "invalid data_source_id" — the positional arg isn't a UUID or collection:// URI. Multi-source database URLs are NOT supported in v0.2.0 — pass the data source UUID directly.

Typical output (after a successful dump)

When reporting results to the user after an enumeration:

1. State the count. If partial: true, flag that the count is incomplete.
2. If relevant, summarize by whatever the user was asking about (domain split, status breakdown, etc.). Use Bash + jq on the output_path file rather than dumping raw rows into the conversation.
3. If the user asked for specifics (e.g. "what are the oldest"), follow up with a targeted slice from the dumped file.

Never flood the conversation with >50 raw rows. The whole point of output_path is to keep bulk data out of context.

See also

• /luke-tasks — routes enumerate-style queries into this tool automatically.
• Notion data-source query reference: https://developers.notion.com/reference/query-a-data-source