Manage, analyze, and optimize your ad campaigns across 7 platforms through natural language in Claude Code and Claude Desktop.
115 expert skills · 35 workflow commands · 5 specialized agents · 33 MCP tools
| Platform | Skills | MCP Tools | What you can do |
|---|---|---|---|
| Meta Ads | 23 | 9 | Campaign management, creative fatigue analysis, audience targeting, ad creation, budget optimization |
| Google Ads | 32 | 4 | GAQL queries, Performance Max, keyword strategy, bid optimization, campaign creation |
| Google Analytics 4 | 18 | 2 | Event tracking, attribution models, audience building, ecommerce analysis |
| Google Search Console | 2 | 3 | SEO performance, keyword rankings, indexing diagnostics, sitemap management |
| Google Tag Manager | · | 3 | Container auditing, consent mode, server-side tagging, conversion setup |
| LinkedIn Ads | 11 | 4 | B2B lead gen, ABM targeting, revenue attribution, Thought Leader Ads, CTV |
| TikTok Ads | 10 | 6 | Creative fatigue, video performance, Smart+ campaigns, Search Ads, Shop |
| Cross-platform | 19 | 2 | Unified workflow + skill tools; attribution reconciliation, SEO vs SEA gaps, budget allocation, competitive analysis |
Pick the path that matches how you use Claude. All four give you the MCP tools; only Claude Code and Cowork give you the bundled skills, commands, and agents.
Requires Claude Code 2.1.74 or newer. Earlier versions had an upstream bug (anthropics/claude-code#21560) that prevented plugin-defined subagents from accessing MCP tools; it is fixed in 2.1.74+. Upgrade with npm i -g @anthropic-ai/claude-code@latest (terminal) or via the VSCode extension marketplace, then Developer: Reload Window.
Add the marketplace once, then install the plugin:
/plugin marketplace add Ad-Superpowers/ad-superpowers-plugin
/plugin install ad-superpowers@ad-superpowersThat's it. The MCP server, all 115 skills, 35 workflow commands, 5 agents, and safety confirmations are bundled.
Cowork supports plugins through an admin UI, not through a slash command. This path is available on Team and Enterprise plans, and only Owners or Primary Owners can add plugins for the organization.
- In Claude Desktop, go to Organization settings → Plugins
- Click Add plugin and choose Sync from GitHub
- Enter the repository:
Ad-Superpowers/ad-superpowers-plugin - Set distribution: Available for install (members can opt in) or Installed by default
- Members then open Browse plugins inside Cowork to install it
Once installed, members get the skills, commands, and MCP tools inside Cowork. The 5 bundled subagents do not currently launch in Cowork — see Known limitations below for details and the Skill-based workarounds.
The regular Claude Desktop chat interface doesn't have a plugin system. You can still get the MCP tools by adding Ad Superpowers as a Connector, but you won't get the bundled skills, commands, or agents (those need Claude Code or Cowork).
- Open Claude Desktop
- Go to Settings → Connectors → Add Custom Connector
- Enter the URL:
https://mcp.adsuperpowers.ai/v1 - Complete the OAuth flow in your browser
If you want the full experience, use Claude Code or Cowork instead.
ChatGPT supports the Ad Superpowers MCP server through the Apps Directory. Search for "Ad Superpowers" in the ChatGPT Apps interface and authorize access. You'll get the MCP tools but not the bundled skills, commands, or agents.
- Create an account at app.adsuperpowers.ai
- Connect your ad platforms through OAuth in the dashboard
- Start asking Claude about your campaigns
No API keys to configure, no manual MCP setup. Authentication happens via OAuth through the Ad Superpowers dashboard.
- "How are my Meta campaigns performing this week?"
- "Which Google Ads keywords have the highest CPA?"
- "Compare my LinkedIn and Meta lead gen performance"
- "Is my TikTok creative showing signs of fatigue?"
- "What's my organic vs paid keyword overlap?"
| Command | What it does |
|---|---|
/ad-superpowers:weekly-performance-pulse |
Cross-platform weekly health check |
/ad-superpowers:creative-fatigue-scanner |
Detect fatigued creatives across platforms |
/ad-superpowers:budget-allocation-planner |
Data-driven budget reallocation recommendations |
/ad-superpowers:keyword-opportunities |
SEO vs paid keyword gap analysis |
/ad-superpowers:monthly-executive-summary |
Executive-ready monthly performance report |
/ad-superpowers:anomaly-detective |
Detect performance anomalies across campaigns |
/ad-superpowers:competitive-landscape-analyzer |
Competitive positioning analysis |
Claude automatically delegates to the right agent based on your task:
| Agent | Focus | Triggers |
|---|---|---|
| marketing-strategist | Strategy, competitive analysis, channel selection, personas | "analyze competition", "which channels?", "new client onboarding" |
| media-buyer | Campaign execution, budgets, bidding, audiences, scaling | "create campaign", "optimize budget", "scale winners" |
| creative-analyst | Creative fatigue detection, A/B tests, refresh strategy | "check creative fatigue", "which ads are tired" |
| reporting-agent | Structured performance reports and comparisons | "weekly report", "monthly summary" |
| tracking-specialist | GTM audits, conversion tracking, SEO vs SEA analysis | "audit my GTM", "keyword overlap", "missing conversions" |
Claude automatically uses the right skill based on your question. Skills provide decision frameworks, benchmarks, and actionable recommendations for:
- Campaign optimization (bid strategies, budget scaling, learning phase management)
- Creative analysis (fatigue detection, A/B testing, diversification)
- Audience strategy (lookalike planning, overlap detection, remarketing)
- Measurement (attribution models, conversion tracking, CAPI setup)
- Technical work (GAQL queries, GTM auditing, GA4 event tracking)
- Create Meta ads (image upload, creative, and ad in one step)
- Duplicate ads with creative overrides
- Pause or resume campaigns and update budgets
- Create Google Ads campaigns (budget, campaign, ad group, ads, keywords)
All write operations include safety confirmations:
- Campaigns always create paused so you can review before going live
- Budget changes require explicit approval
- Destructive operations are clearly flagged
The 5 bundled agents (marketing-strategist, media-buyer, creative-analyst, reporting-agent, tracking-specialist) launch correctly in Claude Code (terminal and VSCode extension, 2.1.74+). They do not currently launch in Cowork, the agentic coding mode inside Claude Desktop.
What's happening: When the plugin is installed in a Cowork session, Cowork's plugin loader reads the plugin's skills, commands, and MCP server correctly, but it does not register the markdown files in agents/ with the Agent tool's subagent_type registry. The Agent tool in Cowork still only exposes its five built-in subagents (general-purpose, Explore, Plan, statusline-setup, claude-code-guide). Attempting Agent(subagent_type="creative-analyst", ...) returns "agent type not found." Our agent files validate cleanly against the plugin spec, so this is a Cowork harness limitation, not a plugin defect.
This is consistent with other known Cowork plugin-loading issues tracked upstream (#39274, #40774), and there is currently no plugin-author-side opt-in, whitelist, or setting that changes this behavior. We have filed a request with Anthropic to expose plugin-bundled agents through Cowork's Agent tool.
What works in Cowork today: Everything except Agent() dispatch.
- All 115 skills load and auto-invoke on relevant questions
- All 35 workflow commands are discoverable and runnable
- All 33 MCP tools are callable from the main conversation
- All write operations (create campaign, duplicate ad, update budget) work
- The agents' prompts and logic are also available as slash commands and skills — you get the same analysis, it just runs in the main context window instead of a separate subagent
So in practice, asking Claude directly in Cowork — "run a creative fatigue scan on my Meta account" or "write a weekly performance report" — gives you the same deliverable the subagent would have produced. It loads the relevant skill (meta-creative-fatigue-analyzer, weekly-performance-pulse, etc.) and calls the MCP tools directly in the main thread.
No known limitations on 2.1.74+. All skills, commands, MCP tools, and subagents work as documented. Earlier versions had #21560 (subagents couldn't access plugin MCP tools); that is fixed.
| Component | Claude Code CLI / VSCode | Cowork (Claude Desktop) | Claude Desktop (chat, via Connector) | ChatGPT (Apps) |
|---|---|---|---|---|
| 115 skills | ✅ | ✅ | ❌ | ❌ |
| 35 commands | ✅ | ✅ | ❌ | ❌ |
| 33 MCP tools | ✅ | ✅ | ✅ | ✅ |
| 5 subagents | ✅ | ❌ (use skills instead) | ❌ | ❌ |
| Tier | Monthly | Annual | Tool calls/mo | Accounts | Users | Platforms |
|---|---|---|---|---|---|---|
| Trial | €0 (14-day) | — | 200 | 25 | 1 | All 7 |
| Pro | €99 | €79 | 1,500 | 25 | 1 | All 7 |
| Team | €249 | €199 | 6,000 | 100 | 5 | All 7 |
| Enterprise | Custom | Custom | Unlimited | Unlimited | Unlimited | All plus custom integrations |
Trial requires a credit card and converts to Pro after 14 days unless cancelled. Annual plans get a 20% discount. A "tool call" is one MCP tool invocation; workflows that fan out to multiple tools count each invocation. Start at adsuperpowers.ai.
- OAuth authentication, credentials are never stored locally
- All API communication over HTTPS
- Users can revoke platform access at any time
- Per-user authentication in team environments
- See SECURITY.md for our responsible disclosure policy
- Website: adsuperpowers.ai
- Dashboard: app.adsuperpowers.ai
- Email: contact@adsuperpowers.ai
MIT, see LICENSE