feat: apify actor generate-schema — AI generates INPUT_SCHEMA.json from existing Actor code #1441

Problem

INPUT_SCHEMA.json is the most tedious part of publishing an Apify Actor. Developers typically:

1. Write their Actor code with hardcoded or process.env-style inputs
2. Manually inspect the code to figure out what inputs exist
3. Hand-write INPUT_SCHEMA.json from scratch, often getting field types or required flags wrong
4. Iterate because apify validate-schema or the platform build rejects it

This friction prevents many Actors from ever getting published with a proper schema.

Proposal

Add a new command that reads your Actor code and generates a draft INPUT_SCHEMA.json:

apify actor generate-schema

What it does

1. Reads src/main.js (or main.py, src/main.ts, etc.) from the current Actor directory
2. Extracts input references: input.url, Actor.getInput(), process.env.APIFY_*, argparse args, etc.
3. Sends the relevant code snippets to an LLM with a schema generation prompt
4. Writes a draft .actor/INPUT_SCHEMA.json
5. Runs apify validate-schema on the result and reports any issues

Example

$ apify actor generate-schema

Reading src/main.js...
Found input references:
  input.startUrl    (string, used in fetch call)
  input.maxPages    (number, default 10)
  input.proxyConfig (object, passed to ProxyConfiguration)

Generating INPUT_SCHEMA.json...
  ✓ Written to .actor/INPUT_SCHEMA.json

Validating...
  ✓ Schema is valid

Review .actor/INPUT_SCHEMA.json and adjust descriptions before pushing.

Generated schema (example)

{
  "title": "My Actor",
  "description": "Auto-generated schema — review before publishing",
  "type": "object",
  "schemaVersion": 1,
  "properties": {
    "startUrl": {
      "title": "Start URL",
      "type": "string",
      "description": "The URL to start scraping from",
      "editor": "textfield"
    },
    "maxPages": {
      "title": "Max pages",
      "type": "integer",
      "description": "Maximum number of pages to scrape",
      "default": 10
    }
  },
  "required": ["startUrl"]
}

Why it matters

• Schema creation is the single biggest friction point before publishing
• Removes the need to know the full JSON Schema + Apify schema extensions spec
• Increases the number of Actors published with proper schemas (better discoverability, better MCP tool descriptions)
• Natural complement to apify create --from-url (feat: apify create --from-url <website> — AI generates Actor scaffold from a target URL #1440) and apify actors expose-as-mcp (feat: expose any Actor as an MCP tool — apify actors expose-as-mcp #1439)

Open questions

• Which LLM / API to call (Apify-hosted vs user brings API key)?
• Support for Python (Actor.get_input(), typer, argparse)?
• Should it update an existing schema or only create from scratch?

Related

• feat: apify create --from-url <website> — AI generates Actor scaffold from a target URL #1440 — apify create --from-url (AI scaffold generation)
• feat: expose any Actor as an MCP tool — apify actors expose-as-mcp #1439 — expose Actor as MCP tool (uses INPUT_SCHEMA for tool description)
• apify validate-schema passes schemas the platform build rejects #1364 — apify validate-schema passes schemas the platform rejects (validation gap)
• generate-schema-types: support directory path as argument #1033 (closed) — generate-schema-types (type gen from schema, not schema from code)