Note
Updated May 2026. Marketing skills, MCPs, and CLIs for AI agents stopped being optional in 2026. agentskills.io became an open standard with nearly 40 clients on the official showcase including Claude Code, OpenAI Codex, Cursor, Gemini CLI, VS Code, GitHub Copilot, Goose, Letta, Kiro, OpenHands, Factory, Snowflake Cortex Code, and Databricks Genie Code. Hermes Agent (Nous Research) and OpenClaw implement the same SKILL.md format outside the official client carousel. Composio's awesome-claude-skills repo curated 1,000+ skills across categories. Hyper shipped a public skills repo. This is the operator guide to the 10 best marketing skills, MCPs, and CLIs across the major agent platforms.
If you installed Claude Code, Codex CLI, Hermes Agent, or OpenClaw to run marketing tasks, the question in May 2026 is which combination of skills plus MCPs plus CLI host turns the agent from a chatbot into a working ad operator. The trust gap is real: ClawSecure's February-March 2026 audit of 2,890+ skills found that 41.7% contained substantive security vulnerabilities, with 30.6% rated high or critical severity (covered by eSecurity Planet and VentureBeat). The discovery gap is real: operators on Reddit and X regularly post about installing five overlapping skills and watching the agent pick the wrong one. The integration gap is real: most marketing-agent setups still require stitching together Composio plus Apify plus Firecrawl plus custom plumbing for Meta Ads, Google Ads, Klaviyo, Shopify, and the rest. This guide ranks the 10 best marketing skills, MCPs, and CLIs operating across all the major agent platforms, with the cross-CLI argument explicit because skills are an open standard now and the Claude-only assumption is already out of date.
Why marketing skills matter in 2026
The line that captures the shift is a Twitter quote from late April 2026: "Most people still use Claude like a chatbot. He turned it into an operating system." The "he" is anyone who installed find-skills, layered in a marketing skill pack from VoltAgent or Composio's awesome-claude-skills repo, connected an MCP server like Hyper or Composio's Rube, and gave the agent permission to act on Meta Ads, Google Ads, Shopify, Klaviyo, and GA4 in real campaigns.
That shift is why the search volume on "claude code skills" hit 3,000 monthly searches and "openclaw skills" hit 4,200 in early 2026. Operators stopped asking whether AI agents could run marketing and started asking which skills, MCPs, and CLI hosts to install. The answer changes a lot depending on what you already use, which platforms you run paid spend on, and how much trust you place in the skill ecosystem.
Skills, MCPs, and CLIs explained
Three distinct layers, often conflated in operator threads:
-
Skills are markdown files with YAML frontmatter that describe a capability. The SKILL.md format started in Claude Code, was open-sourced as agentskills.io, and is now adopted by 40+ products. A marketing skill might look like "30-day-multi-platform-scrape" or "meta-ads-fatigue-audit". Skills live in repos; agents pull them as needed.
-
MCPs (Model Context Protocol servers) are the integration layer. An MCP server like Hyper's connects an AI agent to Meta Ads, Google Ads, Shopify, Klaviyo, GA4 (plus 75 more integrations in Hyper's case) through one OAuth flow. MCP servers expose tools that skills (and the agent itself) can call.
-
CLIs are the agent hosts where you install skills and connect MCPs: Claude Code (Anthropic), Codex CLI (OpenAI), Hermes Agent (Nous Research), OpenClaw (open-source), plus Cursor, Gemini CLI, VS Code with Copilot, Goose, Letta, Kiro, and others.
A working marketing-agent setup combines all three: one CLI as the host, one or two MCPs for integrations, plus a curated set of skills for specific workflows. The 10 entries below cover each layer, weighted toward the combinations that actually work for marketing teams in 2026.
How we ranked these
Six axes that matter for marketers:
- Marketing coverage. Does it actually solve marketing problems (paid ads execution, analytics, CRM, retention, creative) or is marketing one category among many?
- Cross-CLI compatibility. Does it work in Claude Code, Codex CLI, Hermes, OpenClaw, Cursor, Gemini CLI, plus the rest? Or is it locked to one host?
- Trust and audit posture. Audited skills, scoped permissions, OAuth-gated writes vs unscoped community uploads with no review.
- Discovery quality. How does the agent find the right skill, especially when the user has installed dozens?
- Setup time. From signup to first useful query (2 minutes for hosted vs 1-3 days for custom builds).
- 12-month cost. Free OSS vs subscription tiers vs usage-based pricing across realistic operator volumes.
The 10 best marketing skills and MCPs
Marketing teams running paid ads, analytics, retention, and reporting through an AI agent across multiple platforms with audited skills plus scoped permissions
- Best for
- Marketing teams running paid ads, analytics, retention, and reporting through an AI agent across multiple platforms with audited skills plus scoped permissions
- Pricing
- Free 30-day trial, then 49 USD/month
Pros
- The only vertical-marketing integration gateway (Composio, Mastra, Pica are all horizontal)
- Covers Meta Ads, Google Ads, TikTok, Amazon, Pinterest, LinkedIn, plus 75+ adjacent integrations (Klaviyo, GA4, GSC, GTM, Shopify, HubSpot, Ahrefs) in one MCP
- Cross-CLI compatible: works in Claude Code, Codex CLI, Hermes Agent, OpenClaw, Cursor, Gemini CLI
- Public skills repo (open-source) with curated, audited marketing skills scoped to single integrations
- Every skill OAuth-gated with declared permissions; no agent can spend, post, or pixel-fire outside scope
- Built-in rate limiting prevents the 30+ changes per hour pattern that flags ad accounts
- Real customer outcomes: 1,000+ customers, 10M+ USD/month managed ad spend
- Approval workflows for stakeholder review before ads ship
Cons
- Paid (49 USD/month flat) vs free official single-platform MCPs
- Marketing-vertical focus; not the right pick for general engineering or sales workflows
- Best ROI for accounts running 5K USD/month or more across multiple paid channels
Engineering-comfortable operators who want broad horizontal integration coverage plus access to the most-cited open-source skill registry
- Best for
- Engineering-comfortable operators who want broad horizontal integration coverage plus access to the most-cited open-source skill registry
- Pricing
- Free 20K calls/month, then 29 USD/month (200K) or 229 USD/month (2M)
Pros
- 28,100+ GitHub stars on the main repo. Strong DR signal for both the product and the curated skill list
- 1,000+ tools across 250+ services covered by Composio's MCP and HTTP toolkit
- Maintains awesome-claude-skills (the most-cited curated skill list with an explicit Business and Marketing category)
- Rube MCP exposes 500+ apps through one server (one OAuth, many integrations)
- Pay-per-use pricing scales cleanly with agent activity
- Compatible with Claude, ChatGPT, Cursor, Codex, plus more agent clients
Cons
- Horizontal coverage; marketing is one category of many vs Hyper's vertical-marketing focus
- More HTTP toolkit than agent-based marketing execution (raw API exposure rather than intent translation)
- Free tier caps at 20K tool calls per month; multi-platform marketing operators exceed this quickly
- Less specialized for paid-media-specific workflows than Hyper or Pulselane
Marketing operators who want the most widely-adopted agent host with the deepest native Skills integration and the largest community library
- Best for
- Marketing operators who want the most widely-adopted agent host with the deepest native Skills integration and the largest community library
- Pricing
- Free CLI; Claude API usage billed separately
Pros
- Most-installed agent CLI for marketing operators per Reddit and X scrape data
- Native Skills system with find-skills built in for agent-driven discovery
- Largest community skill library (1,000+ skills curated across awesome-claude-skills repos)
- Deep MCP support; works with Hyper, Composio, official Meta Ads MCP, official Google Ads MCP, plus more
- Sub-agent support enables purpose-built marketing roles (paid agent plus retention agent plus analytics agent)
- Strong cross-platform install pattern (Mac, Linux, Windows)
Cons
- No native marketing skills; the host expects you to assemble the marketing stack from external repos and MCPs
- Context window blow-up is a real operator complaint when multiple heavy MCPs are connected at once
- Skills discovery still requires the user (or an agent skill like find-skills) to pick correctly
- Anthropic API usage costs add up at heavy marketing-execution volume
Operators who want the largest skill marketplace plus the second-most-active community after Claude Code, with caveats around skill audit posture
- Best for
- Operators who want the largest skill marketplace plus the second-most-active community after Claude Code, with caveats around skill audit posture
- Pricing
- Free (open-source); third-party paid skill packs available
Pros
- Largest community skill marketplace; 4,200 monthly searches on 'openclaw skills' as of early 2026
- Open-source; no vendor lock-in
- agentskills.io compatible; SKILL.md files transfer cleanly to Claude Code, Codex, Hermes
- VoltAgent's awesome-agent-skills curates 1,100+ entries with explicit marketing collections
- Strong app-store-style discovery via clawhub and SkillsMP indexes
Cons
- Security audit through early 2026 surfaced vulnerabilities in roughly 41% of marketplace skills
- No central audit or review process; operator responsible for vetting skills before install
- Discovery noise: hundreds of overlapping marketing skills with inconsistent quality
- Marketing-specific skills lag Claude Code community quality
Marketers using OpenAI models as the agent backbone who want first-party support for agentskills.io and growing MCP integration
- Best for
- Marketers using OpenAI models as the agent backbone who want first-party support for agentskills.io and growing MCP integration
- Pricing
- Free CLI; OpenAI API usage billed separately
Pros
- Official OpenAI CLI with first-party model support
- agentskills.io compatible (adopted by Codex in early 2026)
- Growing MCP ecosystem; Hyper, Composio, official Meta Ads MCP, official Google Ads MCP all supported
- Cross-platform install (Mac, Linux, Windows)
- Strong terminal-native operator UX
- Apify, Firecrawl, plus other marketing-adjacent MCP integrations work out of the box
Cons
- Smaller community skill library than Claude Code or OpenClaw as of May 2026
- Newer entrant; ecosystem maturing fast but lags the established CLIs
- Marketing operators report sub-agent ergonomics behind Claude Code's Task tool
- OpenAI API usage costs at heavy execution volume
Engineering-led marketing teams that want to build custom agent workflows on an open-source framework with cloud observability
- Best for
- Engineering-led marketing teams that want to build custom agent workflows on an open-source framework with cloud observability
- Pricing
- Free framework; Cloud from 250 USD/month for teams
Pros
- 23,800+ GitHub stars on the main repo. Strong DR signal for the framework
- Apache 2.0 framework supports 40+ model providers (Claude, OpenAI, plus more)
- Mastra Cloud handles observability, memory, plus deployment at the 250 USD/month team tier
- Y Combinator W25 backing; stable funding posture
- Strong agent orchestration primitives for multi-step marketing workflows
Cons
- Framework not product; you build the marketing-specific workflows yourself
- Engineering-required; not a one-click marketing operator setup
- Marketing-vertical features must be built from generic agent primitives
- 250 USD/month cloud tier is a step up from Hyper's flat 49 USD/month
Marketers who want a curated bundle of marketing-specific skills (SEO, CRO, paid, retention, content, pricing) installable as a unit
- Best for
- Marketers who want a curated bundle of marketing-specific skills (SEO, CRO, paid, retention, content, pricing) installable as a unit
- Pricing
- Free (open-source skill pack)
Pros
- 34-skill bundle covering SEO and content, copywriting, CRO, email and lifecycle, paid advertising, pricing and monetization, product marketing, sales enablement, plus analytics and RevOps
- Created by Corey Haines (founder, well-known marketing operator) with operator-grade specificity
- Referenced in VoltAgent's awesome-agent-skills repo (1,100+ entries total)
- Cross-CLI compatible per the agentskills.io spec
- Free; install via Claude Code, OpenClaw, Codex, Hermes, plus more hosts
Cons
- Skills cover diagnosis and analysis; execution (actually launching ads, sending emails) still requires an MCP integration
- No built-in MCP server; works alongside Hyper, Composio, or official platform MCPs
- Single maintainer; update cadence varies
- Documentation lighter than enterprise products
Open-source-first operators who want a free agent host with automatic skill selection and growing momentum in the marketing-skill community
- Best for
- Open-source-first operators who want a free agent host with automatic skill selection and growing momentum in the marketing-skill community
- Pricing
- Free (open-source agent)
Pros
- Open-source agent with auto-skill selection (the agent picks the right skill from the installed set automatically)
- Strong recent momentum; operator threads through April 2026 favor Hermes over OpenClaw for skill UX
- agentskills.io compatible; pulls from the same community skill libraries
- Free; no per-seat pricing for operator self-host
- Nous Research is a known open-source AI research org; reputation premium
Cons
- Smaller community skill library than Claude Code or OpenClaw
- Self-host required for full control; hosted options limited
- Marketing-specific skill availability still light compared to top hosts
- Newer entrant; long-term stability less established than Anthropic or OpenAI hosts
Operators who installed more than 10 skills and want the agent to pick the right one automatically rather than guessing
- Best for
- Operators who installed more than 10 skills and want the agent to pick the right one automatically rather than guessing
- Pricing
- Free (open-source skill)
Pros
- Viral discovery skill: install once, describe the outcome you want, the system picks the right skill from hundreds available
- Solves the most-cited operator pain (agent picks the wrong skill from an overlapping installed set)
- Cross-CLI compatible per agentskills.io spec
- Free; install via Claude Code, OpenClaw, Codex, Hermes
- Strong community momentum through Q1 and Q2 2026
Cons
- Not a marketing skill itself; meta-skill that selects other skills
- Quality of recommendation depends on the underlying skill library quality
- No integration layer; you still need MCPs for actual marketing execution
- Multiple forks exist; pick one carefully and pin the version
Marketers who need a 30-day intelligence sweep across Reddit, X, YouTube, TikTok, plus HackerNews on any topic in one skill invocation
- Best for
- Marketers who need a 30-day intelligence sweep across Reddit, X, YouTube, TikTok, plus HackerNews on any topic in one skill invocation
- Pricing
- Free (open-source skill)
Pros
- Single-purpose skill that pulls 30-day mentions across Reddit, X, YouTube, TikTok, plus HackerNews
- 25,400+ stars on GitHub; one of the most-installed standalone skills in 2026
- Cross-platform monitoring built in (saves hours of manual scrape work per week)
- Cross-CLI compatible per agentskills.io
- Free; install via Claude Code, OpenClaw, Codex, Hermes
Cons
- Single-purpose (intelligence and monitoring, not execution)
- API costs for the underlying scrape services add up at heavy use
- No native integration with paid ads or CRM tools (read-only intelligence layer)
- Best paired with execution MCPs like Hyper or Composio
Side-by-side comparison
| Tool | Layer | Marketing focus | Cross-CLI | Pricing |
|---|---|---|---|---|
| Hyper MCP plus Skills | MCP plus Skills | Vertical marketing (paid, analytics, retention, reporting) | Yes | 49 USD/mo |
| Composio | MCP plus Skills curation | Horizontal (marketing is one category) | Yes | Free 20K calls then 29 USD/mo |
| Claude Code | CLI host | Generalist host; deepest community skill library | Native host | Free CLI plus API costs |
| OpenClaw | CLI host plus marketplace | Generalist host; largest skill marketplace | Native host | Free |
| Codex CLI | CLI host | Generalist host; growing MCP ecosystem | Native host | Free CLI plus API costs |
| Mastra | Agent framework | Build your own marketing workflows | Yes | Free plus 250 USD/mo Cloud |
| Corey Haines Marketing Skills | Skill bundle | Marketing-specific (32 skills) | Yes | Free |
| Hermes Agent | CLI host | Generalist host with auto skill selection | Native host | Free |
| Find Skills | Meta-skill | Discovery layer (any vertical) | Yes | Free |
| last30days-skill | Single skill | Cross-platform monitoring | Yes | Free |
Trust: skills with vulnerabilities
A recurring operator complaint through Q1 and Q2 2026: skill marketplaces are noisy, lightly-curated, and increasingly targeted by malicious uploads. ClawSecure's February-March 2026 audit of 2,890+ OpenClaw skills surfaced that 41.7% contained substantive security vulnerabilities, 30.6% rated high or critical severity, and 99.3% shipped without a config.json permissions manifest. Two named-malware incidents (Remcos RAT and GhostLoader variants) shipped through legitimate-looking community skill uploads before being pulled.
This is the gap Hyper, Composio, and a small set of curators are trying to close. Three patterns that signal trust:
-
Audited, scoped skills. Every skill is open-source, readable, scoped to a single integration, and declares its permissions explicitly. Operators can audit before installing.
-
OAuth-gated writes. Skills cannot spend, post, or pixel-fire outside the declared OAuth scope. An ad-launch skill cannot also read your inbox.
-
Maintained-by-vendor signal. Skills maintained by a company with skin in the game (Hyper for marketing, Composio for general integrations) are reviewed before publication and rolled back on issue reports faster than community-uploaded skills.
The pragmatic operator move in 2026: install community skills for monitoring and intelligence (low blast radius) but route execution-grade workflows (ad spend, email sends, customer data writes) through audited, vendor-maintained MCPs and skill packs.
How to find the right skill
The most-cited operator pain on Reddit and X through 2026: installing five overlapping skills and watching the agent pick the wrong one. The find-skills community skill addresses this with auto-selection (install it once, describe the outcome you want, the system picks the right skill from hundreds available). The Hyper skills repo addresses it with curation (one canonical skill per integration, no overlap, agent picks correctly by default).
Three patterns that work in practice:
-
Install fewer skills, install better ones. Three audited, well-scoped marketing skills beat 30 overlapping community uploads. The agent makes fewer wrong-skill choices when the option set is curated.
-
Use find-skills (or equivalent) as the meta layer. Lets the agent introspect the installed set and pick correctly without you having to remember which skill does what.
-
Tag skills by job function in your SKILL.md. "paid-ads.meta.create-campaign" beats "meta-skill". The find-skills selection layer (and the agent's own context) gets better signal when the skill name is operator-readable.
Skills are cross-CLI now
The agentskills.io spec became an open standard through early 2026. The SKILL.md format that started in Claude Code now runs natively across nearly 40 client products on the official showcase: OpenAI Codex, Cursor, Gemini CLI, VS Code, GitHub Copilot, Goose, Letta, Kiro, OpenHands, Factory, Snowflake Cortex Code, plus Databricks Genie Code. Hermes Agent and OpenClaw implement the same SKILL.md format outside the official client list, which means skills installed in one host work across the others. Operators do not need to pick a CLI host first and then find skills for it; they can pick skills and install them across whichever host they prefer.
This matters for marketing teams because most marketing-operator setups span more than one host. A senior marketing operator might use Claude Code for strategic analysis, Codex CLI for content generation, Hermes Agent for monitoring (auto-skill selection is strong here), plus OpenClaw for community-contributed marketing-skill experimentation. The skills installed in each host can be identical because the spec is shared.
The operator implication: pick skills and MCPs that are cross-CLI compatible, not host-specific. Hyper, Composio, Mastra, plus all the top community skill packs (find-skills, last30days, Corey Haines marketing bundle) are all cross-CLI by design.
Hyper vs Composio, Mastra, and Pica
Four products often grouped together as "AI agent integration gateways." The honest comparison:
| Dimension | Hyper | Composio | Mastra | Pica |
|---|---|---|---|---|
| Focus | Vertical marketing | Horizontal (sales plus dev plus marketing) | Open-source agent framework | Horizontal connector library |
| Public count | 80+ marketing integrations | 1,000+ tools across 250+ services | 40+ model providers | 250+ platforms |
| Skills repo | Public, marketing-vertical | awesome-claude-skills (1,000+ skills) | Framework primitives | OSS deprecated, moved to cloud |
| Pricing | Free 30-day trial, then 49 USD/month flat | Free 20K calls, then 29 USD/mo (200K) | Free framework, 250 USD/mo Cloud | Usage-based |
| MCP server | Yes (Hyper MCP) | Yes (Rube MCP) | Yes (via framework) | Yes (Pica MCP) |
| Best for | Marketing teams running paid ads, analytics, retention | Engineering-built workflows across many verticals | Custom agent builds | Connector breadth |
The simple test: if marketing is the bottleneck (paid ads execution, analytics, retention, agency reporting), Hyper. If you need horizontal integration coverage across sales, support, and general engineering, Composio. If you want to build custom agent workflows on a framework, Mastra. If you need raw connector breadth without marketing specialization, Pica.
These are peers, not direct competitors. The integration-gateway category is large enough that vertical and horizontal players coexist.
How to choose for your stack
Match your marketing-agent stack to your operation
Use Hyper MCP plus Skills if
Recommended: You run paid ads, analytics, retention, or reporting through an AI agent. You spend 5K USD/month or more across multiple platforms. You want audited, scoped, OAuth-gated skills plus an agent execution layer for Meta Ads, Google Ads, TikTok, Amazon, Pinterest, LinkedIn, Klaviyo, Shopify, GA4, GSC, HubSpot, plus 70+ more. You want flat 49 USD/month pricing.
Use Composio plus awesome-claude-skills if
Recommended: You need horizontal integration coverage across sales, support, engineering, and marketing. You are comfortable wiring up HTTP toolkits and writing custom workflows. You want pay-per-use pricing that scales with agent activity. You want access to the most-cited curated skill list (1,000+ skills, explicit marketing categories).
Use Claude Code plus community skills if
Recommended: You want the most-installed agent CLI plus the largest community skill library. You are willing to assemble a marketing stack from external MCPs plus skill packs yourself. You accept the discovery and audit overhead of unscoped community skills.
Use OpenClaw plus skill marketplace if
Recommended: You want the largest skill marketplace plus open-source freedom. You will personally audit skills before installing (or use Hyper/Composio for execution-grade workflows and OpenClaw for low-risk monitoring or intelligence skills only).
Use Codex CLI or Hermes Agent if
Recommended: You prefer the OpenAI agent ecosystem (Codex) or want auto-skill-selection from an open-source host (Hermes). Both are cross-CLI compatible per agentskills.io and work with Hyper, Composio, plus the major community skill packs.
Use Mastra if
Recommended: You have an engineering team building custom agent workflows. You want an open-source framework with cloud observability and memory primitives. You accept the 250 USD/month Cloud tier for production deployment.
How Hyper helps marketing teams
Hyper sits at #1 in this ranking because the marketing-vertical lane is genuinely uncovered by the horizontal integration gateways. Composio, Mastra, and Pica are excellent products for the surface they target (horizontal integration coverage), but none of them are built around the specific shape of a marketing operator's day: launching Meta Ads campaigns, diagnosing Google Ads spend leakage, reconciling platform-reported ROAS against GA4 data, generating retention email variants in Klaviyo, plus reporting cross-platform performance to stakeholders.
What Hyper does specifically:
-
One MCP, 80+ marketing integrations. Meta Ads, Google Ads, TikTok, Amazon, Pinterest, LinkedIn ads plus Klaviyo, GA4, GSC, GTM, Shopify, HubSpot, Ahrefs data, social posting (Instagram, TikTok, LinkedIn) all in one OAuth flow.
-
Audited, scoped skills. Every skill in the public Hyper skills repo is open-source, readable, scoped to a single integration, and declares its permissions explicitly. No agent can spend, post, or pixel-fire outside the declared scope.
-
Built-in rate limiting. Prevents the 30+ changes per hour pattern that flags ad accounts. Hyper translates AI intent into compliant API calls rather than exposing raw API endpoints.
-
Cross-CLI by design. Works in Claude Code, Codex CLI, Hermes Agent, OpenClaw, Cursor, Gemini CLI per the agentskills.io standard.
-
Real customer outcomes. 1,000+ customers, 10M+ USD/month managed ad spend, documented case studies at /blog/ai-marketing-case-study.
The pragmatic stack for most marketing operators in May 2026: Claude Code or Codex CLI as the host, Hyper MCP for marketing integrations plus skills, find-skills for discovery on top of any additional community skills installed, plus selective community skills (last30days for monitoring, Corey Haines bundle for diagnostics) routed to read-only roles.
Autonomous marketing
Grow your business faster with AI agents
- Automates Google, Meta + 5 more platforms
- Handles your SEO end to end
- Improves website conversions
- Runs social media for you
Frequently asked questions
Q: What are AI agent skills and how are they different from MCPs?
Skills are markdown files with YAML frontmatter that describe a capability the agent can use. The SKILL.md format started in Claude Code, was open-sourced as agentskills.io, and is now adopted by 40+ products (Codex CLI, Hermes Agent, OpenClaw, Cursor, Gemini CLI, plus more). MCPs (Model Context Protocol servers) are the integration layer: an MCP server like Hyper connects an agent to Meta Ads, Google Ads, Klaviyo, GA4, plus 75 more services through one OAuth flow. Skills describe capabilities. MCPs provide integrations. A working marketing setup uses both.
Q: How do I make Claude Code (or OpenClaw, Hermes, Codex) actually run my Google Ads or Meta Ads?
Install an MCP that connects the agent to the ad platforms. For multi-platform paid execution (Meta plus Google plus TikTok plus Amazon), Hyper MCP at 49 USD/month with the free 30-day trial covers all of them in one OAuth flow with built-in rate limiting and audited skills. For Meta-only operators, the official Meta Ads MCP (April 29 2026 launch) is free. For Google-only operators, the official Google Ads MCP (April 28 2026 launch) is free but read-only as of May 2026. Most marketing operators run 2+ platforms and converge on multi-platform MCPs.
Q: How do I find the right marketing skill out of hundreds available without my agent picking the wrong one?
Three approaches. (1) Use find-skills (the community meta-skill that auto-selects from your installed set based on outcome description). (2) Install fewer, better-audited skills rather than dozens of overlapping community uploads. Hyper's skills repo ships one canonical skill per integration with no overlap. (3) Tag skills with job-function-readable names in your SKILL.md so the agent's selection layer gets cleaner signal (e.g. 'paid-ads.meta.create-campaign' beats 'meta-skill').
Q: How do I trust a marketing skill enough to give it ad-spend permissions?
Three trust signals to verify. (1) Open-source and readable: you should be able to inspect the skill code before installing. (2) Scoped permissions: the SKILL.md should declare what OAuth scopes and API endpoints it touches, and the host should enforce that scope. (3) Maintained by an accountable party: skills maintained by a company with skin in the game (Hyper for marketing, Composio for general integrations) are reviewed and rolled back faster on issues than community uploads. ClawSecure's February-March 2026 audit of 2,890+ OpenClaw skills flagged 41.7% with substantive vulnerabilities (30.6% rated high or critical), which is why vendor-maintained marketing skills became the preferred route for execution-grade workflows.
Q: Are marketing skills cross-CLI or do I have to pick a platform?
Cross-CLI is now the standard. agentskills.io became an open standard through early 2026 and is adopted by 40+ products including Claude Code, Codex CLI, Hermes Agent, OpenClaw, Cursor, Gemini CLI, VS Code with Copilot, Goose, Letta, Kiro, plus OpenHands. The SKILL.md format that started in Claude Code now runs natively in all of these. Skills you install in one host work in another. The same applies to MCP servers: Hyper, Composio, official Meta Ads MCP, plus official Google Ads MCP all work across the major CLI hosts.
Q: How does Hyper compare to Composio for marketing teams?
Different focus. Composio is horizontal: 1,000+ tools across 250+ services covering sales, support, engineering, and marketing as one category among many. Hyper is vertical-marketing: 80+ integrations specifically for paid ads, analytics, retention, reporting, plus creative. For marketing teams running paid spend across multiple platforms, Hyper's marketing-specific skills, audit posture, and agent execution layer pay back vs Composio's horizontal HTTP toolkit. For engineering teams building cross-vertical workflows, Composio's breadth pays back. Many operators run both: Composio for general integrations, Hyper for the marketing execution layer.
Q: Are there really 1,000+ marketing skills available?
More than 1,000 skills exist across all categories in awesome-claude-skills and awesome-agent-skills repos; marketing-specific skills number in the low hundreds across discovery, diagnosis, generation, and execution categories. Corey Haines' VoltAgent bundle alone covers 32 marketing-specific skills (SEO, CRO, paid, retention, content). Hyper's public skills repo ships a curated marketing set with one canonical skill per integration. The pragmatic operator pattern is to install 5 to 15 well-curated marketing skills rather than every community upload, since discovery quality drops fast above 20+ installed skills.
Q: What is the best CLI for marketing operators in 2026?
Claude Code has the largest install base and deepest skill ecosystem; most marketing operators converge there as the primary host. Codex CLI is the right pick for OpenAI-model-first operators. Hermes Agent has strong auto-skill-selection and growing momentum. OpenClaw has the largest marketplace but the noisiest curation. Most operators run more than one host across different workflows since agentskills.io makes skills cross-compatible. The CLI matters less than the MCPs and skills you install on top of it.