Ryze marketing-skills: 49 Claude skills that read your ad accounts through an MCP connector

What the marketing-skills repository actually ships

This is a prompt and instruction pack for Claude, not an application. The README describes 49 free skills across four folders: Google Ads, Meta Ads, SEO, and AI visibility. The Google Ads folder also covers cross-platform Google plus Meta work. The bundled MCP server is not part of the repository. It is a hosted endpoint at connector.get-ryze.ai, operated by Ryze AI, the company that also publishes the get-ryze.ai blog linked from the README. The repository name is irinabuht12-oss/marketing-skills and the default branch is main. No release has been published, so the only distribution channel is the repository itself and the zip file linked in the README. The audience is narrow and clear: performance marketers and SEO people who already use Claude and want structured audit workflows rather than a blank chat box.

The two-part mechanism: skills describe the method, MCP supplies the data

The design splits method from data. Each skill is a set of instructions telling Claude how to approach a task, for example finding wasted spend, checking creative fatigue, or building a programmatic page template. The data arrives through the Ryze MCP connector, which the README says gives Claude live Google Ads, Meta Ads, GA4 and Search Console data. When you ask "why did my Meta CPA go up this week", the skill supplies the diagnostic sequence and the connector supplies the numbers. That separation is the interesting part. It also means the skills degrade in a specific way: without the connector, they still run, but only against exports you paste in by hand. The README states this directly in the requirements section, and it is the single most important operational fact about the project.

Installing the plugin, the connector, or neither

The README gives three installation paths. The plugin route is two commands: claude plugin marketplace add irinabuht12-oss/marketing-skills, then claude plugin install marketing-skills@ryze. The manual route clones the repository and copies the skill folders into your Claude skills directory: git clone https://github.com/irinabuht12-oss/marketing-skills && cp -r marketing-skills/skills/* ~/.claude/skills/. The connector is added separately with claude mcp add ryze --transport http https://connector.get-ryze.ai/mcp. In claude.ai or Claude Desktop the README points to Settings, then Connectors, then Add custom connector, pasting the same URL. It also states the connector works in Cursor and ChatGPT. Note the ordering claim in the README: the plugin route says the connector comes bundled, while the manual route does not mention bundling at all. If you take the manual path, expect to add the MCP endpoint yourself.

Where the skill set is thin and where it is specific

The skill table mixes genuinely specific work with generic marketing prompts. Google Ads Audit, Meta Ads Audit and the A/B Test Analyzer name concrete outputs: search term leaks, negative keyword gaps, audience overlap, statistical significance, sample size. Those are the skills that justify the repository. Others, such as Email Sequence Writer and Content Repurposer, are prompt templates that would work with any capable model and any data source, connector or not. The README lists 49 skills in the repository while the linked guide advertises 98. That gap is not explained in the README, and it is worth resolving before you build a workflow around a specific skill name. The AI visibility folder is the most differentiated item, since it targets citation in ChatGPT, Claude, Gemini, Perplexity and AI Overviews, a question set that most marketing prompt packs do not address at all.

The connector is the real dependency, and the real risk

Every skill that depends on live account data depends on a hosted service you do not control. The endpoint is https://connector.get-ryze.ai/mcp, and the README describes it as free. Free hosted connectors change terms, rate limits and availability on the vendor's schedule, not yours. The repository contains no fallback data layer, no local cache, and no documented export format that the skills are guaranteed to parse. The README says the skills are MIT licensed, but the repository has no LICENSE file and GitHub records no licence. For a prompt pack that is less alarming than for a library you link against, but it matters if you plan to fork the skills, embed them in an internal tool, or redistribute them to a team. Treat the missing LICENSE file as an open question.

How this differs from wiring Claude to a warehouse or a BI tool

The obvious alternative is to skip the skill pack and connect Claude to your own data path, for example exporting ad data into BigQuery and querying it through a database MCP server you host yourself. The difference is control versus setup. A self-hosted path means you own the credentials, the schema and the retention policy, and you write the diagnostic prompts from scratch. This repository inverts that: the prompts come pre-written and the data path is someone else's service. A second alternative is the platform's own tooling. Google Ads and Meta Ads both ship recommendation engines inside their interfaces, and those run on first-party data with no connector in between. They will not accept a custom prompt, and they will not produce a cross-platform budget reallocation view the way the Ad Spend Allocator skill claims to. Pick based on whether you want pre-written method with a third-party data hop, or full control with more assembly work.

Maintenance cost and what the repository does not tell you

Skills are text, so they do not break at runtime the way a library does. They go stale instead. Ad platform interfaces, metric definitions and benchmark expectations change, and a skill written against last year's Meta reporting model will produce confident, outdated advice. The repository has no release history, so there is no changelog to check for updates to individual skills. Your upgrade path is re-cloning or re-running the plugin install command and comparing the skills directory yourself. The MCP endpoint can change behaviour without any change to the repository, which means the version you tested and the version running next month may differ. If you need a stable, auditable artifact, copy the skill files into your own repository at a pinned commit and accept that you are now maintaining them.

Editorial conclusion

Adopt this if you already pay for Claude Pro, Max, Team or Enterprise and want audit-style prompts for Google Ads, Meta Ads, LinkedIn Ads, Reddit Ads and SEO without writing them yourself. Do not adopt it if you cannot connect a third-party MCP endpoint to your live ad accounts, or if you need a LICENSE file in the repository before shipping anything derived from these files. Before installing, open the repository and confirm two things: whether a LICENSE file exists, and whether the skills you plan to rely on run on pasted exports alone or assume the Ryze connector is present.

Frequently asked questions

Official sources

1. irinabuht12-oss/marketing-skills on GitHub
2. Issues
3. Project website
4. README