06-reference

zack igclaims meta ads mcp angle mining

2026-05-13·reference·source: X (long-form article)·by Zack (@ig_claims)
meta-adsmcpangle-miningpaid-adssquarelyproductization-gapsanity-check-candidatefde-asymmetric-edge

"how i'm finding winning ad angles with claude + meta MCP" - @ig_claims

Why this is in the vault

Framing correction from founder (2026-05-13 14:36 ET): the value is NOT "install the Meta Ads MCP for Squarely." The value is the workflow SHAPE we adapt to our own surfaces. Zack's piece happens to be a Meta Ads worked example; the discipline transfers to any account-data-meets-creative-decision surface.

Treat as a reference workflow we adapt. Direct hit on the FDE asymmetric-edge productization-gap thesis filed yesterday (solo operator ships enterprise-grade work at $5k-$30k engagement size because MCP-integration removes the bespoke-engagement overhead). Sponsor disclosure is real (lead-magnet CTA at end) but doesn't kill the substance.

⚠️ Sponsorship

Self-promotional. The CTA at the bottom routes to "DM me ANGLES at @ig_claims" for done-for-you service. Author is selling. That said, the workflow is substantive enough to be useful independent of buying his service - he names specific client numbers, the prompt sequences, and the toolset limits. Treat the impressive case studies (especially the $180k/month uplift) with the standard "marketer-sourced, no audit" discount, but the workflow shape itself is real.

The core argument

Meta's official Ads MCP launched 2026-04-29 and gives Claude direct read-and-write access to the Marketing API surface. The obvious use case is campaign management. Zack focuses on what he calls "angle mining" - using the MCP for competitive research + creative-brief generation rather than budget juggling.

Setup (his claim, ~5 min): claude.ai → settings → integrations → add Meta Ads MCP → OAuth via Business Manager credentials. Once connected, 29 tools available via natural language.

3-step workflow:

  1. Account audit - Pull 90 days of campaign data, identify top performers by ROAS + CTR, identify fatigued creatives (CTR declining 3+ consecutive days, frequency above 5), find under-spend gaps (proven angles sitting at low spend that haven't been scaled). Surfaces what's working internally before looking outside.
  2. Competitive landscape map - Manually browse Meta Ads Library, feed competitor angles into Claude as a structured list, ask Claude to identify: which competitor angles overlap with yours (saturated), which you haven't tested (opportunity), which NO competitor is running (blue ocean). Output: 5 ranked new angles to test with one-line hooks each.
  3. Creative brief generation - For top 3 recommended angles, generate hook variations (different emotional triggers), recommended format (founder video / data viz / testimonial / UGC), copy length, audience segment, landing page pair, and predicted performance range based on similar angles in account history.

14-day cycle: he runs steps 1-3 on a 14-day repeating cadence per client - day 1-2 audit, day 3-4 landscape, day 5-7 briefs, day 8-14 test, day 15 restart with fresh data. Claims 20-30 new creative concepts tested per month per client.

Specific impact numbers (marketer-sourced, no audit)

The last one is the load-bearing impact claim. Take with the usual marketer-attribution discount; the workflow shape is the durable artifact.

The abstracted workflow shape (the actual takeaway)

Per founder framing correction: the durable pattern Zack demonstrates abstracts to a 4-step shape independent of Meta Ads specifically:

  1. Pull structured data from an account or surface (paid ad performance, web analytics, email engagement, etc.)
  2. Identify what's working / fatiguing / gap - find sustained winners, declining performers, under-spent proven angles
  3. Cross-reference against the competitive landscape - which of our angles overlap with competitors (saturated), which they're running that we haven't tested (opportunity), which nobody is running (blue ocean)
  4. Generate a data-informed brief for the next test cycle - hooks, format, audience, predicted performance range

The cadence: 14-day cycle (pull → analyze → brief → test → restart). Steps 1-3 produce inputs; step 4 produces the artifact; steps 5-14 are execution; day 15 repeats.

This is the SAME shape as the multi-agent skill pipeline shipped 2026-05-12, applied to marketing/content decisions:

Pipeline seat Angle-mining role
pipeline-spec-author What angles are we considering? (intent + ICP context)
pipeline-test-author What's the eval shape? (CTR, conversion, ROAS, open rate, share count, etc.)
pipeline-code-author Produce the creative brief artifact
pipeline-critic Rank by differentiation + probability against the competitive landscape

Worth adding "angle-mining" as a candidate per-domain config for the multi-agent pipeline (alongside MAC, SC-article, website-discovery from ticket #3). The SC-article config overlaps partially - angle-mining is the broader pattern; SC-article is a domain-specific instantiation.

Mapping against Ray Data Co

Where the workflow shape applies across active RDCO surfaces (independent of Meta Ads MCP install):

  1. Squarely Amazon Ads (existing Notion task 352f7d49-36d1-816e-abed-e2c2304e08a2). No Amazon MCP yet, but the workflow shape transfers. Pull Amazon Ads reports manually, run audit / landscape / brief sequence in Claude. Direct adaptation of Zack's 4-step shape to a non-MCP-enabled surface.

  2. Squarely Meta Ads (existing Notion task 352f7d49-36d1-81c4-b2c5-e81a8b1d548b). The MCP install is parked - founder explicitly NOT wiring up now (no cycles, plus Squarely has zero Meta Ads spend so audit step would produce nothing). Workflow shape comes online when both conditions resolve: spend exists + founder cycles available for OAuth.

  3. Squarely website analytics-driven decisions. Once the website BLOCKER PR fixes the Cloudflare Web Analytics CORS issue (S3 in the audit), the workflow shape applies: page-level engagement as account data, other puzzle/book sites as competitive landscape, monthly cycle.

  4. Sanity Check content angle mining. Substack engagement (open rate, click-through, shares) = the account data. Other data-engineering newsletters = the competitive landscape. 14-day editorial cycle = cadence. This is the highest-friction-to-lowest-friction transition because we already have the SC voice + the audit-by-issue discipline; we'd just be adding the data layer. Most-immediate-fit application of Zack's shape.

  5. MAC content angle mining (longer-horizon). Less mature - we don't have engagement data yet because MAC hasn't shipped at scale. When MAC content starts running, the same shape applies.

FDE asymmetric-edge concept article (filed 2026-05-13 02:50 ET) - this is the worked example. Zack operates exactly in the band we positioned RDCO for: $5k-$30k engagement size, productized artifact (creative brief) is the deliverable, SMB clients big FDE shops won't touch. He's running the playbook the concept article predicted would crystallize once productization-via-MCP removed the enterprise-only floor. The MCP enables the solo operator to ship enterprise-grade angle mining at SMB pricing because the bespoke-engagement overhead (analyst time + tooling stack + integration custom work) collapses to one OAuth click + a Claude session.

Pattern-recognition / Jaynit / RLHF-shaped pipeline. Angle mining IS pattern recognition applied to ad creative. The 14-day cycle IS the tight-feedback-loop Jaynit prescribed (volume + tight feedback + articulation). The MCP is the cheap-consultation layer; the chiddush is the angle synthesis. Founder critique of selected angles → preference labels for the next cycle's brief generation. Same RLHF-shape we mapped to the multi-agent skill pipeline last night, applied to paid-ads creative.

Sanity Check candidate (parked, no-derivative discipline). Trap: "MCPs are cool, here's a workflow." That's derivative restatement of Zack. The ORIGINAL reframe: "the productization gap meets the MCP-enabled solo operator." Substance:

This is parked, not pitched. If founder wants it landed, it goes alongside the FDE concept article in the positioning cluster.

Caveats worth flagging

Related