meta-ads-drive-to-meta

name

meta-ads-drive-to-meta

description

End-to-end orchestrator — take a creative sitting in Google Drive and turn it into a ready-to-launch (PAUSED) Meta/Facebook ad, in one guided flow. Use when the user says "sube este creativo de Drive a Meta y ármame el anuncio", "from Google Drive to Meta Ads", "publica (en pausa) este video/imagen de mi Drive", or wants the whole chain (asset → campaign → ad set → ad) from a Drive file without stitching the sub-skills by hand. It chains meta-ads-asset-upload → meta-ads-campaign-setup → meta-ads-adset-setup → meta-ads-ad-setup, adds the Drive-source resolution + transport decision that make it CONSISTENT, verifies, and (for tests) tears down. Everything is created PAUSED; activation is a human decision.

Meta Ads — Drive → Meta (end-to-end orchestrator)

One job: a file in Google Drive → a PAUSED, ready-to-review Meta ad, reliably, for a non-technical user. This skill does not re-document each step — it sequences the existing skills and owns the two things that make the chain break: getting the Drive bytes into Meta (size/transport decision) and wiring the ids together (account blob, page, creative hash, budget units).

🔒 Safety contract (inherited): everything is created PAUSED; activation is human. account_id for every MCP call = the signed blob from list_accounts, passed verbatim (one wrong character → account_ref signature is invalid). Budgets are minor units (×currency offset) with a per-account minimum the API enforces. See references/pipeline.md.

The chain (what runs, in order)

1. Resolve account + page — list_accounts (signed blob) + discover the Page id.
2. Asset from Drive → hash/id — ../asset-upload/ + scripts/drive_to_meta_upload.py.
3. Campaign (PAUSED) — ../campaign-setup/.
4. Ad set (PAUSED, targeting + budget) — ../adset-setup/.
5. Ad (PAUSED, creative = the Drive hash + copy + CTA) — ../ad-setup/.
6. Verify the ad's creative == the uploaded hash; report ids. (Tests: tear down with campaign_delete.)

The full ordered call sequence, params, and every validated gotcha live in references/pipeline.md — read it before running.

What makes it CONSISTENT (the differentiator)

• Transport decision is automatic, not guessed (the #1 source of failures):
  • Drive file ≤ 10 MB → pull bytes with the Google-Drive MCP (download_file_content) → transport B: prepare_upload + scripts/drive_to_meta_upload.py (base64 POSTed in code, never streamed through the model).
  • Drive file > 10 MB (all real videos) → the Drive MCP cannot download it → transport A: the file must be link-shared and passed as a public url to facebook_ads.video_upload. (Claude does NOT change Drive sharing — the user does; see pipeline.md.)
• Currency-aware budget: read the account currency, convert the user's real money to minor units, enforce the account minimum — the user always thinks in "$X/day", never in cents.
• Video is async: after video_upload, poll object_read(video_id, fields="status") until ready before ad_create, or Meta rejects the ad.

Inputs (what to collect from the user)

Input	Needed for	Notes
Ad account (name or id)	all	Resolved via list_accounts → signed blob.
Drive file (id or name)	asset	An image (JPG/PNG — WebP is rejected) or video (MP4/MOV).
Objective	campaign	OUTCOME_TRAFFIC / LEADS / SALES / AWARENESS / ENGAGEMENT / APP_PROMOTION.
Facebook Page	ad	Auto-discoverable: object_read(act_…, fields="promote_pages{id,name}"). Confirm with the user.
Budget (real money/day)	ad set	Converted to minor units; ≥ account minimum.
Geo (≥1 country)	ad set	Meta rejects targeting with no geo.
Copy: message / headline / link / CTA	ad	LEADS also needs a lead_gen_form_id.

Operate (happy path, image)

"Sube business-retail.jpg de mi Drive a la cuenta Porter y ármame un anuncio de tráfico a portermetrics.com, $5/día, US, en pausa."

1. list_accounts(query="Porter") → signed blob. object_read(act_…, "promote_pages{id,name}") → Page.
2. Drive file ≤10 MB → download_file_content → prepare_upload(image_upload) → python3 scripts/drive_to_meta_upload.py --kind image --from-drive-json --src <dl.json> --account act_… --upload-url <fresh> --filename business-retail.jpg --mime image/jpeg → image_hash.
3. campaign_create(objective=OUTCOME_TRAFFIC, special_ad_categories=[], status=PAUSED).
4. adset_create(optimization_goal=LINK_CLICKS, billing_event=IMPRESSIONS, destination_type=WEBSITE, targeting_countries=["US"], daily_budget_amount=<minor≥min>, status=PAUSED).
5. ad_create(page_id, image_hash, message, headline, link, cta_type=LEARN_MORE, status=PAUSED).
6. object_read(ad_id, "creative{image_hash},status") → confirm hash matches + PAUSED. Report ids.

Validated end-to-end live 2026-07-16 (account Porter act_794709130739347, COP): Drive JPG → hash 3b21ca91… → campaign → ad set (min budget 3319 COP minor) → ad (creative == hash) → verified PAUSED → deleted. See references/pipeline.md for the video path + edge cases.

Scope / boundary

• ✅ Orchestrates: Drive source → asset id → campaign → ad set → ad (PAUSED) → verify → (test) teardown.
• ❌ Does NOT re-implement targeting/objective/creative-format detail — that lives in the sub-skills (campaign-setup, adset-setup, ad-setup, asset-upload). This skill decides the order, the transport, and the id plumbing.
• ❌ Does NOT activate ads, change Drive sharing, or invent budgets/audiences.