What a Claude skill is vs a prompt
A prompt is something you type each time. A skill is a small package Claude keeps on hand and loads when it decides the task calls for it. Anthropic defines a skill as a folder containing a SKILL.md file that begins with YAML frontmatter holding two required fields, a name and a description, followed by instructions in Markdown. A skill can also carry reference files and scripts.
The important part is how Claude uses the description. Anthropic calls it progressive disclosure: Claude sees only the name and description of every enabled skill, which costs very little, and reads the full instructions only when your request matches. Ask "why did conversions drop last week?" and a skill whose description covers diagnosing performance drops is loaded, without you naming it. In Claude Code you can also call a skill directly by typing a slash and its name.
So the difference is not length or cleverness. A skill is worth making when three things are true: you do the task repeatedly, you want it done the same way each time, and the method is specific enough to write down, such as your thresholds for wasted spend, the order you check things in, or the structure of a client report. A one-off question is still best as a prompt.
- Prompt: typed each time, lives in one conversation, depends on you remembering every step.
- Skill: installed once, loaded automatically when relevant, identical across conversations and team members.
- Connector (MCP server): gives Claude access to live systems such as Google Ads. Skills and connectors are separate and work together.
Installing a skill in Claude and Claude Code
The steps below follow Anthropic's Help Center and Claude Code documentation as checked on 25 September 2026. Skills are available on Free, Pro, Max, Team and Enterprise plans, and in the Claude apps they need code execution switched on.
In the Claude apps (web and desktop): on Free, Pro and Max plans, open Settings, Capabilities and make sure Code execution and file creation is on. Then go to Customize, Skills, click the plus button, choose Create skill, then Upload a skill, and upload a ZIP. The ZIP must contain the skill folder itself at its root, and the folder name should match the skill's name. The skill then appears in your list with a toggle. On Team plans skills are on by default; on Enterprise plans an owner must first enable skills and code execution under Organization settings, Plugins and skills. On both, owners can provision a skill for the whole organisation.
In Claude Code: create a folder under ~/.claude/skills/ for skills you want in every project, or under .claude/skills/ inside a repository to share them with everyone who works in it, and put the SKILL.md file in it. The folder name becomes the command, so ~/.claude/skills/weekly-client-report/SKILL.md can be run as /weekly-client-report. Claude Code watches those folders and picks up new or edited skills in the current session without a restart. If you sign in to Claude Code with your claude.ai account, skills from your claude.ai account sync across too.
Converting a skill from our library. Each skill in the MCP Ads library downloads as a single Markdown file with frontmatter, for example google-ads-search-terms-mining.md. It is short on purpose: a name, a description of what the skill checks and delivers, and the phrase that triggers it. To install one, create a folder named after the name field, save the file inside it as SKILL.md, and place the folder in ~/.claude/skills/ or ZIP it for upload. Anthropic's Help Center gives a 200-character limit for an uploaded skill's description and several library descriptions are longer, so if an upload is rejected, cut the frontmatter down to name and a shorter description. Then make it yours: add your CPA or ROAS targets, date ranges and report format. As Anthropic advises, review any downloaded skill before enabling it.
Google Ads skills
Six skills that cover the work a Google Ads account needs every week, from a first audit to the report. Each links to its category page in the library, where it can be downloaded on its own or with the rest of its category.
- 1. Google Ads Full Account Audit (google-ads-account-audit). For a new client or a quarterly review: campaign structure, wasted spend, quality scores, bidding strategy, conversion tracking, ad copy and negative keywords, ending in a prioritised fix list. Trigger: "Run a full Google Ads audit for [account]". In Audits and health.
- 2. Wasted Spend Finder (google-ads-wasted-spend-finder). Finds keywords spending with zero conversions, irrelevant search terms triggering ads and campaigns burning budget with no return. Trigger: "Find all wasted spend in [account]". In Audits and health.
- 3. Search Terms Mining (google-ads-search-terms-mining). Classifies every query in the search terms report: converting terms to add as keywords, high-spend irrelevant terms to add as negatives, and match type changes. In Keywords.
- 4. Search Term to Negative Converter (google-ads-search-term-to-negative). Paste a list of irrelevant search terms and it formats them as negatives at the right match type and campaign level. In Negative keywords, next to the industry negative lists and the conflict check that catches negatives blocking good keywords.
- 5. RSA Builder (google-ads-rsa-builder). Writes complete Responsive Search Ads, 15 headlines and 4 descriptions, following Google's guidance on keyword inclusion and benefit-led copy. Trigger: "Write RSA ads for [product/service/campaign]". In Ads and creative.
- 6. Weekly Google Ads Report (google-ads-weekly-report). This week against last on spend, clicks, impressions, CTR, conversions, conversion rate, CPA and ROAS per campaign, with highlights. In Reporting.
Meta Ads skills
Meta work is less about queries and more about audiences, creative and frequency, and the skills reflect that.
- 7. Meta Ads Full Account Audit (meta-ads-account-audit). Campaign structure and objective alignment, audience overlap between ad sets, creative fatigue, pixel health, budget distribution and placements. Trigger: "Run a full Meta Ads audit for [account]". In Audits and setup.
- 8. Creative Fatigue Check (meta-ads-creative-fatigue-check). Flags ads with frequency above 2.5 as a warning and above 3.5 as danger, alongside CTR and CPA trends, so you know which creative to replace. In Audits and setup, with the audience overlap and pixel audits.
- 9. Creative Brief (meta-ads-creative-brief). Uses what has worked in the account to define the next ads: format (image, video, carousel, reel), the hook, the key message and the call to action. In Audiences and creative.
- 10. Weekly Meta Ads Report (meta-ads-weekly-report). This week against last on reach, impressions, frequency, clicks, CTR, conversions, CPA, ROAS and spend, with best and worst ads. In Meta reporting.
- 11. Meta vs Google Attribution (meta-vs-google-attribution). Both platforms claim credit for the same sales under their own attribution windows; this skill uses GA4 as the independent reference to show where sales are really coming from. In Meta reporting.
GA4 and SEO skills
Ad platforms grade their own homework. These skills bring in Google Analytics 4 and Search Console so decisions rest on what the site actually recorded.
- 12. GA4 Attribution Truth (ga4-attribution-truth). Compares the conversions Google Ads and Meta report with what GA4 recorded, to size the over-reporting from overlap and view-through counting. In GA4 traffic and funnels.
- 13. Landing Page Performance (ga4-landing-page-performance). Ranks landing pages that receive paid traffic by conversion rate and flags pages wasting ad spend or whose conversion rate fell this week. In GA4 traffic and funnels. Lead generation businesses should also look at the GA4 lead generation skills.
- 14. SEO Keyword Opportunities (seo-keyword-opportunities). Finds queries ranking in positions 8 to 20 in Search Console, where modest work can reach page one, plus high-impression, low-CTR queries. In SEO content and rankings.
- 15. SEO CTR Optimisation (seo-ctr-optimisation). Finds pages ranking in the top 10 with a click-through rate below what their position should earn, and rewrites the title and meta description. In SEO content and rankings.
Agency routines
The last five are not about one platform. They set context and run the recurring checks an account manager does for every client, which is where skills save the most time.
- 16. Client Setup (client-setup). Run first. Records the business type, what it sells, monthly budget, target cost per sale or lead and the platforms in use, so every other skill judges a lead-gen client on cost per lead and an eCommerce client on ROAS. In Foundation skills.
- 17. Daily Account Pulse (daily-account-pulse). A morning check of Google Ads and Meta for disapproved ads, budget problems, delivery drops and tracking failures. Trigger: "Run the daily pulse for [client]". In Daily workflows.
- 18. Weekly Agency Routine (agency-weekly-routine). The Monday workflow for one client in sequence: account health, search terms, bids, ad review, Meta creative fatigue and SEO. In Agency management.
- 19. Pre-Launch Checklist (pre-launch-checklist). Before any campaign goes live: conversion tracking, budgets, targeting, ad approvals, UTM parameters, negative keywords and audience exclusions. In Client reporting and management.
- 20. Performance Drop Diagnosis (performance-drop-diagnosis). For when conversions or leads suddenly fall: works through Google Ads, Meta Ads, GA4 and Search Console to find the cause. In Foundation skills.
Connecting skills to live ad accounts (why skills need an MCP server)
A skill tells Claude how to do an audit. It does not give Claude the numbers. Install the Wasted Spend Finder without a connection and Claude can explain the method or work on a CSV you paste in, but it cannot open your account, and anything it says about "your" search terms is guesswork. Anthropic's own guidance makes the same split: use skills for the workflow and MCP connections for external services.
An MCP server is that connection. Once one is connected, Claude can call tools that read campaign, keyword, search term, audience and creative data from the ad platforms, and, with permission, make changes. The skill decides the order and the judgement; the server supplies the facts.
MCP Ads is the server the skills in our library are written for. It connects Google Ads, Meta Ads, GA4, Search Console and Tag Manager, plus TikTok, LinkedIn, Microsoft, Snapchat, X and ChatGPT Ads, to Claude, Claude Code, ChatGPT, Cursor and other clients, and several Google and Meta logins can be connected at once for agencies. Changes wait for your approval, and campaigns are created paused. Read-only skills such as audits and reports never write anything. The Google Ads MCP server and Meta Ads MCP pages list what each platform supports.
If you do not want to install anything, you can still use the library: connect MCP Ads and paste a skill's trigger phrase, or its description, into the chat. Claude will do the task with live data, but from its own judgement rather than the saved procedure, so results vary more between runs. Installing the skill is what makes the tenth run look like the first.
For the wider picture of where Claude fits in marketing work, read Claude for Marketing: The Complete Guide. Agencies running many client accounts should start with MCP Ads for agencies.
