Your AI assistant can already work with RightMessage. Connect it through our MCP server, and it can read your project, draft a survey, or set up a personalized headline.
But knowing which buttons exist was never what made a setup work. The hard part is deciding what to ask your visitors, what to change based on their answers, what to do with that data afterward, and how to check that the right person actually sees the right thing.
That’s what we’ve spent eight years helping RightMessage customers figure out. So we wrote it down.
RightMessage Skills are nine free playbooks your AI assistant can follow when you ask it to audit your account, segment new leads, build a quiz, personalize a page, or decide what to test next. They’re public on GitHub, and you can use them whether or not your assistant supports skills.
Access was never the hard part
We’ve talked with a lot of customers who had RightMessage installed and weren’t sure what to do next. Their questions were rarely about which setting to click. They sounded more like this:
- What should we ask new subscribers, and what do we do with the answer?
- Should this quiz end with a product recommendation, a booked call, or both?
- Why are customers still seeing an offer for something they already bought?
- The preview looks right. Why isn’t it showing up on the live site?
The same handful of problems kept coming back. Answers collected and never used. Quiz results that didn’t lead anywhere. A preview mistaken for proof that the live site works. Rules based on ad names that broke the next time someone renamed a campaign. Success measured in clicks instead of the sale or booking you actually wanted.
An AI assistant with access to your account can make every one of those mistakes, just faster.
So we took the approach we use with customers, plus the fixes for those recurring problems, and turned them into playbooks your assistant can follow. They’re procedures, not customer transcripts or private account details.
They won’t make your assistant an expert on your business. They make it ask better questions before it touches anything.
What a skill is, and how it works with MCP
A skill is a written playbook your AI assistant consults for a particular job. It’s a folder containing a SKILL.md file: a short description of when to use it, followed by step-by-step instructions and checklists, sometimes with a template. The format is an open standard called Agent Skills.
Your assistant doesn’t read every skill all the time. It keeps each skill’s short description in mind, then loads the full instructions when your request matches, like “build a quiz” or “why isn’t my widget showing up?”
MCP, short for Model Context Protocol, is the connection that lets your assistant work inside your RightMessage account with your permission. Through the RightMessage MCP server, it can read your projects and results, draft changes, and return previews.
Put simply, MCP gives your assistant access. A skill tells it how to approach the job.
When your task involves building something, the skills follow this general workflow:
- Understand what you sell and who you’re trying to reach.
- Reuse the data and setup you already have.
- Agree on one useful change before building anything.
- Build it as a draft and read it back.
- Test the whole path, not just the preview.
- Publish only when you say so.
That last step matters. Configuration changes, like a new survey or campaign, stay in draft until someone publishes them. Other actions take effect right away, like starting a test or creating a field in your email platform, so the skills tell your assistant to ask you first. Audits and analysis can stop at a recommendation without changing anything.
They’re still instructions, not permission controls. Your assistant’s approval settings and the access you grant it are what actually limit it.
Nine skills, one for each job
You don’t need all nine, and you don’t need an audit before every small change. Start with the job you already have.
- Account audit looks at your business, your website, and your RightMessage project, then recommends the three most useful next moves and a plan for the first one. It doesn’t change anything by default. Try: “Audit this project and its website. What are the three most useful improvements? Don’t change anything yet.”
- MCP operations helps you connect, confirm your assistant is working in the right project, troubleshoot a missing widget, review a draft, and publish or undo a change. Try: “Preview works but the live widget is missing. Find the first unverified step before changing anything.”
Learn what your visitors need
- Segmentation helps you decide what to ask new or existing leads, where their answers get saved, and what should happen next because of them, like a survey on your thank-you page. Try: “Build a post-opt-in survey that guides new leads to the right next step. Reuse our existing CRM fields and leave the configuration in draft.”
- Quiz builder designs assessments, product recommendation quizzes, and lead qualifiers backward from the result. Each outcome needs a reason it fits and a next step before any questions get written. Try: “Help me design a product recommendation quiz. Show me the outcomes and question strategy before building it.”
Change what they see
- Smart CTAs show prospects, leads, and customers different next steps, so you stop asking subscribers to join a list they’re already on. Try: “Stop asking known customers to join our mailing list. Draft a better CTA flow with an anonymous fallback.”
- Website personalization adapts the headline, proof, and call to action on a page you already have for audiences you can actually recognize, and keeps the original for everyone else. Try: “Personalize our pricing page for our two main audiences. Keep the original for everyone else and return previews.”
- Paid-traffic message match makes a landing page follow through on the ad that earned the click. It checks real ad URLs first, because a campaign name that works today can change tomorrow. Try: “Check these real ad-click URLs, then propose two message-matched treatments for this landing page. Don’t edit our ads.”
Follow up and improve
- Email personalization turns what you’ve learned into conditional copy for your email platform, with a sensible default when an answer is missing. It writes the email. It doesn’t send it. Try: “Turn these survey answers into a personalized welcome email. Use our actual platform’s syntax and don’t send it.”
- Optimization reads your results, checks that you’re measuring the conversion you care about, and recommends one experiment you can defend. Try: “Is this flow losing people at the opener or after the result? Show the evidence before recommending a test.”
Install them in an assistant that supports skills
If you use Claude Code, Cursor, Codex, or another assistant that supports the Agent Skills standard, install the skills with the open-source Skills CLI. Run this in your terminal:
npx skills add rightmessage/mcp-skills
The installer asks which skills you want and which of your assistants should get them. To install just one:
npx skills add rightmessage/mcp-skills --skill rightmessage-quiz-builder
In Claude Code, you can install the same skills as a plugin instead. Run these inside a Claude Code session:
/plugin marketplace add rightmessage/mcp-skills
/plugin install rightmessage@rightmessage-skills
Once they’re installed, you don’t have to name a skill. Ask for the job, and your assistant picks the matching playbook.
These commands need a terminal or an assistant that runs on your computer. Pasting them into a web chat won’t install anything.
No skills support? Hand your assistant the repository
Plenty of chat assistants can’t install skills. Many can still follow one if they can read the files. Copy this prompt, describe your goal, and paste it into a new conversation:
Review https://github.com/rightmessage/mcp-skills. My goal is [describe your goal]. Read the README, then the relevant SKILL.md file and any supporting files it requires. Use that workflow to help me. Check whether you have a working RightMessage connection and confirm the project before accessing my account. Start with a recommendation; don't change anything yet. If you can't retrieve the files, tell me what to upload.
This only lasts for the current conversation. Unlike an installed skill, you’ll need to paste the prompt again next time.
It also depends on your assistant actually opening the files. Seeing a GitHub link isn’t the same as reading the skill. If your assistant can’t browse, download the relevant SKILL.md and the files it mentions from the repository, then upload them. The prompt asks it to tell you when that’s necessary rather than guess.
Skills don’t connect your account
Whichever route you choose, connecting RightMessage is a separate step. Installing or reading a skill doesn’t connect your account, add the RightMessage script to your site, grant permissions, or approve anything for publishing.
To work inside your account, add the RightMessage MCP server to your assistant, sign in, and confirm which project it should use. The setup guide walks through the common assistants. MCP access is included on every RightMessage plan.
Without that connection, your assistant can still plan from what you tell it, like survey questions or page copy. It just can’t look at your account, and the skills tell it not to pretend it did. Your website builder, email platform, and ad account each need their own connection and permission too.
Start with a question, not a change
New to using RightMessage with an AI assistant? Start with the audit prompt above. You’ll see how your assistant reads your business before it changes anything.
Already know the job? Go straight to that skill. A survey that finally does something with its answers. A page that stops treating customers like strangers. A quiz that ends with an offer that fits.
Not using RightMessage yet? Start a RightMessage trial, connect your assistant, and ask it where to begin.
