Closes #882

What

• get-dataset stops returning the raw Apify schema. The flat fields list it already returns is the complete, projection-ready field inventory.
• Remove the nudges that steer the LLM into get-dataset-schema (a context-heavy call): dropped from get-dataset's description, and from get-dataset-items' last-page nextStep (now points at get-dataset for the field list). This keeps the loaded-tool gating added in [Bug]: Fix abort-actor-run and get-dataset-items from external review #1007 — it just redirects the gated reference from get-dataset-schema to get-dataset.
• get-dataset-schema is unchanged — still available on demand, returns the full inferred schema (no depth cap).

Why

Two findings drove this:

• Probe of the 10 top store Actors (#882 comment): the raw dataset.schema is 93–95% of the get-dataset response bytes on top Actors and declares 23–39% phantom fields (disabled add-ons, etc.). Meanwhile the flat fields list already carries every projectable path, including deep ones (media.videoDeliveryLegacyFields.dash_manifest_url), and is exactly the format get-dataset-items fields= consumes.
• Mixpanel: get-dataset-schema is rarely called. So the win isn't capping its output — it's (a) dropping the redundant raw schema from the tool people do use, and (b) not steering the model into a heavy call it seldom needs. fields is the one canonical description of dataset shape.

The tradeoff: the model no longer sees types/formats up front. It learns them by reading an item (which it does anyway), or by calling get-dataset-schema explicitly.

Effect

get-dataset on a compass/crawler-google-places dataset: 41 KB → 2.9 KB, key set identical minus schema.

{
 "actorSpecification": 1,
 "views": {
  "overview": {
   "title": "Overview",
   "transformation": { "fields": ["title", "totalScore", "reviewsCount", "street", "city", "website", "phone", "url"] },
   "display": { "component": "table", "properties": { "title": { "label": "Place name" } } }
  },
  "leadsEnrichment": {
   "title": "Lead Enrichment",
   "transformation": { "fields": ["title", "firstName", "lastName", "email", "jobTitle", "companyName", "leadsEnrichment"], "unwind": ["leadsEnrichment"] },
   "display": { "component": "table" }
  }
 },
 "fields_excerpt": {
  "orderOnline": {
   "type": "object",
   "properties": {
    "pickUps": { "type": "array", "items": { "$ref": "#/definitions/PickUpItem" }, "description": "Pickup options" },
    "deliveries": { "type": "array", "items": { "$ref": "#/definitions/DeliveryItem" }, "description": "Delivery options" }
   },
   "description": "Online ordering options. Omitted unless `scrapeOrderOnline` is enabled.",
   "examples": [{ "pickUps": [{ "name": "Pickup", "orderUrl": "https://example.com/pickup" }] }]
  }
 }
}

{
 "id": "gUWSHSZ1fALKL1pgB",
 "name": null,
 "userId": "…",
 "itemCount": 3,
 "cleanItemCount": 3,
 "actId": "nwua9Gu5YrADL7ZDj",
 "actRunId": "xoPvwNjdbrkWZUSZH",
 "stats": { "storageBytes": 2711, "readCount": 23, "writeCount": 3 },
 "fields": [
  "title", "categoryName", "address", "city", "countryCode", "phone", "website", "url",
  "totalScore", "reviewsCount", "placeId", "location.lat", "location.lng",
  "openingHours.day", "openingHours.hours",
  "additionalInfo.Service options.Delivery", "additionalInfo.Service options.Dine-in",
  "additionalInfo.Payments.Credit cards"
  /* …85 paths total, all dot-notation, projection-ready */
 ],
 "consoleUrl": "https://console.apify.com/storage/datasets/gUWSHSZ1fALKL1pgB",
 "generalAccess": "FOLLOW_USER_SETTING",
 "summary": "Dataset 'gUWSHSZ1fALKL1pgB' has 3 items, 85 fields.",
 "nextStep": "Use get-dataset-items with datasetId=gUWSHSZ1fALKL1pgB and limit (for example 20) to fetch items."
}

Tests

• Unit: get-dataset-items terminal nextStep points at get-dataset (the [Bug]: Fix abort-actor-run and get-dataset-items from external review #1007 gating tests updated accordingly); get-dataset field normalization unchanged. Full suite: 968 passed.
• E2e via mcpc on a fresh dataset (the authoritative check for the schema drop): get-dataset has no schema key and nextStep mentions only items; get-dataset-items routes to get-dataset; get-dataset-schema still returns the full schema on demand.

Not in scope

For wide datasets the flat fields list is itself large (~307 paths for a Facebook-posts dataset). Capping or making fields navigable is a separate question — worth its own issue.

Note for apify-mcp-server-internal

get-dataset structuredContent no longer carries the schema key — contract suite may need a matching update. Not a tool removal (so not breaking in that sense), but the response shape changed.

https://claude.ai/code/session_01Sf9wACoa9h9y2m2WZ2Sde5