n8n DataTables Admin

n8nac has no commands for DataTable lifecycle (n8nac datatable does not exist as of 2.2.x). The only way to create/seed/drop tables outside the n8n UI is the public REST API. This skill encapsulates the exact endpoints + the curl patterns the PreToolUse hook is configured to allow.

Carve-out scope: Only URLs containing /api/v1/data-tables are allowed past the REST-block. The guard catches curl/wget/urllib/Invoke-RestMethod/requests. against /api/v1 — you cannot dodge it by switching HTTP tool. Anything else (workflows, credentials, executions) still routes through n8nac — do not try to reach those via raw HTTP from here.

Prerequisites

.env must define:

N8N_API_URL=https://your-instance.example.com
N8N_API_KEY=<key from n8n UI → Settings → API>

Source it at the top of every shell invocation:

set -a; source .env; set +a
BASE="${N8N_API_URL%/}/api/v1/data-tables"
AUTH=(-H "X-N8N-API-KEY: $N8N_API_KEY" -H "Accept: application/json")

Polling / loops: to read rows repeatedly, loop the carve-out curl — a curl against /api/v1/data-tables inside a while/for still passes the guard. Do NOT switch to python/urllib/Invoke-RestMethod to poll (the guard now blocks those too), and do NOT read n8nac's internal ~/.n8n-manager/secrets.json for the key — it is undocumented and may change. The key is N8N_API_KEY (n8n UI → Settings → API, or your secret store / Infisical).

API Reference (n8n public API v1)

Routes confirmed against packages/cli/src/public-api/v1/handlers/data-tables/ (master).

Tables

Columns

Rows

filter is n8n's row-filter DSL — typically { "type":"and", "filters":[{"columnName":"id","condition":"eq","value":42}] }. Conditions: eq, neq, gt, gte, lt, lte, like, notLike, isNull, isNotNull.

Recipes

List all tables

set -a; source .env; set +a
curl -s "${N8N_API_URL%/}/api/v1/data-tables" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -H "Accept: application/json" | jq .

Create a table with typed columns

curl -s -X POST "${N8N_API_URL%/}/api/v1/data-tables" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -H "Content-Type: application/json" \
  --data-binary @- <<'JSON' | jq .
{
  "name": "customer_queue",
  "projectId": "<project-id>",
  "columns": [
    { "name": "customer_id", "type": "string" },
    { "name": "status",      "type": "string" },
    { "name": "retries",     "type": "number" },
    { "name": "last_seen",   "type": "date"   }
  ]
}
JSON

Umlaut-safe inline JSON on Windows: Never curl -d '{"name":"Geschäft"}' — use a heredoc as shown (--data-binary @-) or write the JSON to a temp file and pass --data-binary @file.json. See feedback_curl_umlaut_body.md.

Seed rows from a local JSON file

# rows.json: { "data": [ {"customer_id":"abc","status":"new","retries":0,"last_seen":"2026-05-18T00:00:00Z"}, ... ] }
TABLE_ID=<id>
curl -s -X POST "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID/rows" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -H "Content-Type: application/json" \
  --data-binary @rows.json | jq '.data | length'

Read first 50 rows

curl -s "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID/rows?limit=50" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" | jq '.data'

Upsert rows by key

curl -s -X PUT "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID/rows" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -H "Content-Type: application/json" \
  --data-binary @- <<'JSON' | jq .
{
  "filter": { "type":"and", "filters":[{"columnName":"customer_id","condition":"eq","value":"abc"}] },
  "data":   { "status":"done", "retries": 1 }
}
JSON

Delete rows by filter

curl -s -X DELETE "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID/rows" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -G \
  --data-urlencode 'filter={"type":"and","filters":[{"columnName":"status","condition":"eq","value":"obsolete"}]}' \
  | jq .

Drop entire table

curl -s -X DELETE "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" | jq .

Add a column to existing table

curl -s -X POST "${N8N_API_URL%/}/api/v1/data-tables/$TABLE_ID/columns" \
  -H "X-N8N-API-KEY: $N8N_API_KEY" -H "Content-Type: application/json" \
  -d '{"name":"notes","type":"string"}' | jq .

Workflow Integration

Once a table exists, reference it from a workflow via the dataTable node (operations: getRows, insertRows, updateRows, upsertRow, deleteRows). Inside the workflow node, filters use the same DSL — see feedback_n8n_datatable_upsert.md for the filters.conditions[].keyName quirk.

For node-level discovery:

npx n8nac skills node-info dataTable --json

Upsert node shape (THREE parts — all required)

A workflow-node upsert needs filters.conditions (with keyName + condition + keyValue) AND matchingColumns. Miss any one and the upsert matches nothing and always inserts (→ duplicates). n8nac validation is authoritative here.

{
  operation: 'upsert',
  dataTableId: { __rl: true, value: '<table-id>', mode: 'id' },
  filters: {
    conditions: [{
      keyName:  'matching_column',          // (1) which column to match
      condition: 'eq',                       // (2) MUST be set
      keyValue: '={{ $json.matching_column }}', // (3) MUST be set
    }],
  },
  columns: {
    mappingMode: 'autoMapInputData',
    value: {},
    matchingColumns: ['matching_column'],    // MUST mirror the condition key
    schema: [ /* all columns with correct `type` (match LLM/extractor output types) */ ],
  },
}

Usage patterns (when to reach for a DataTable)

• Fan-in store for parallel sub-workflows — each async sub-workflow writes a result row (batch_id, item_idx, status, result); the parent polls COUNT(*) WHERE batch_id=X until it reaches N, then reads + merges. The recommended fan-out/fan-in mechanism — see n8n-autopilot:n8n-orchestration-patterns (Pattern B). Avoids webhook-between-workflows + HMAC pain.
• Idempotency / dedup — on retries, upsert on the natural key (e.g. (batch_id, item_idx)) instead of insert, so a re-run overwrites rather than duplicating.
• Error rows, not silent skips — route a risky node's error output to a DataTable write with status: 'error' so the fan-in count still completes and the failure is visible. Never continueOnFail: true (masks silent failures).
• Cross-run state — small dedup/seen-tables, processing cur