Ask Claude or ChatGPT about the shelf

Two minutes to connect. Then every brand on the shelf, every run time and every stopped ad answers in the chat you already use, with the date it was last checked.

Claude

  1. Settings → Connectors → browse the directory → Adshelf → Connect. Adshelf is listed in Claude’s directory, so there is no URL to paste.
  2. Sign in with your work email (a one-time code, no password). New here? The same box creates your free account.
  3. Approve the connection. Claude gets read access to the shelf for your account and nothing else.

Prefer your own connector URL (desktop app, Settings → Plugins → Add MCP server, type Streamable HTTP)? It is on your account page once you are signed in.

ChatGPT

  1. Turn on Developer mode (Settings → Plugins → Advanced on a Plus or Pro account; a workspace admin does it on Team and Enterprise).
  2. In a web browser, chatgpt.com/plugins → the + button → name it Adshelf → Server URL: your connector URL from the account page.
  3. Start a chat, pick Adshelf under Plugins, and ask.

What to ask

These are the questions early users ask most. Each one maps to a tool the connector exposes, so the answer comes with real ads, real dates and a last-checked stamp.

What did my competitors do this week?my_brands_digest — launched, crossed 30 days, stopped, for your starred brands
What has Grüns kept running longest, and when did you last check?get_brand_strategy — proven winners, format mix, medians, last_checked
Tear down HexClad’s creative strategy and tell me the three things worth borrowing.get_brand_ads + get_brand_strategy
Which angles are proven in other categories that nobody in supplements is running?cross_category_angles — the lanes a whole category leaves empty
Find the angles my competitors keep paying for that I have never run.find_my_gaps — needs your Meta account connected; until then it reads the competitor side
Show me every ad on the shelf that has run more than 90 days with a discount angle.search with min_days and status filters
Add Jones Road to my shelf.add_brand — matches the brand on the public Ad Library and starts storing

Prompts by job

Copy the one that matches the task. Each is written so the answer is bounded by what the shelf actually knows (last-checked dates, stops tracked since).

Monday client note · For each of my starred brands, list what launched, what crossed 30 days and what stopped since last Monday, with dates. One paragraph per brand, then the single most important move across all of them.my_brands_digest(days: 7)
Pitch teardown · Pull <brand>’s five longest-running ads and everything it stopped in the last 60 days. Name the angle each runs, say what the stops tell us, and draft three opening slides. Note the last-checked date and whether stops are tracked for this brand.get_brand_strategy + get_brand_ads(status: stopped)
Creative brief · In <industry>, which angles do the 30-day survivors run, which proven angles from other categories are missing here, and what would a first test look like? Cite one real ad with its run days per recommendation.get_benchmarks(industry) + cross_category_angles(industry)
Hook sweep · Search <industry> for ads past 90 days and give me the ten opening lines, grouped by angle, with run days.search(query, industry, min_days: 90)

Your own agents

The connector is plain MCP over HTTPS, so it works anywhere Claude runs: Claude Code, the Agent SDK, a scheduled task. Agencies use it for the Monday note that writes itself: a task that runs my_brands_digest per client every Monday morning and drops the draft in the client channel. Every tool answer carries last_checked and, per brand, stops_tracked_since, so an agent can refuse to claim what the shelf has not verified.

What it can and cannot see

Connect in two minutes

Free for good on the Weekly Edition. 14 days of everything on the paid plans, no card.

Create your account