Marketing teams adopting AI agents keep hitting the same question: should we use MCPs, CLIs, or Skills?
The answer is all three — and picking the wrong one for the job creates friction where there should be flow.
As Jng Iam put it: “I stopped trying to pick a winner and started letting the context decide.”
Here’s how to make that decision for your team.
The three primitives, explained simply
Skills are documented processes. They encode how your team does a specific thing — your writing style, your campaign review checklist, your reporting methodology — written so an AI agent can follow it. A skill is just a Markdown file. No code, no API, no infrastructure. If you can write a Google Doc, you can write a skill.
CLIs (command-line interfaces) are developer tools that have been around for decades. Tools like git, npm, or gh (GitHub’s CLI) let you interact with services by typing commands in a terminal. AI agents can use these tools natively — they already know how most CLIs work.
MCPs (Model Context Protocol servers) are authenticated connections between AI agents and SaaS tools. They handle OAuth, access control, and give the agent structured access to tools like Google Analytics, HubSpot, or Notion.
When to use each one
Use Skills when the knowledge is in your head
Skills shine when the value lives in how your team thinks, not in connecting to a tool.
Examples:
- Brand voice guidelines — so every AI-generated draft sounds like your company, not a chatbot
- Campaign review checklists — the 12 things your team checks before launching
- Reporting templates — how you structure a weekly performance summary
- Content briefs — the format your writers expect
As Simon Willison noted: “The core simplicity of the skills design is why I’m so excited about it.”
Skills require zero technical setup. You write instructions in plain English, save them as a Markdown file, and any AI agent that supports skills can use them immediately. They’re also portable — the same skill works across Claude Code, Cursor, Windsurf, and other tools.
If you’ve ever wished you could clone your best marketer’s brain and hand it to an AI, skills are how you do it.
Use CLIs when you need speed and composability
CLIs are fast, composable, and need no wrapper or abstraction layer.
Eric Holmes makes the case directly: “CLIs have had decades of design iteration. They’re composable, debuggable, and they piggyback on auth systems that already exist.”
For marketing teams, CLI usage typically looks like:
- Content workflows — using Git to version-control your content, deploy site changes, or manage content pipelines
- Data processing — piping CSV exports through transformation scripts
- Build and deploy — running site builds, deploying landing pages, triggering CI/CD pipelines
- Local automation — shell scripts that chain multiple tools together
CLIs work best for people comfortable in a terminal — and they don’t burn tokens the way MCP servers do.
Use MCPs when the team needs governed access to tools
MCPs solve the problem of giving AI agents authenticated, controlled access to your SaaS stack. This matters most when:
- Multiple people need the same integrations — rather than each person configuring their own API keys, an MCP server handles auth centrally
- Non-technical team members need tool access — MCPs abstract away the complexity of API calls
- You need audit trails and access control — MCPs provide a governance layer that raw API calls don’t
- Background agents need to run autonomously — agents processing data overnight need reliable, authenticated connections
If your team uses tools like Google Analytics, HubSpot, Notion, or PostHog, MCP servers let your AI agents query and act on those platforms without anyone writing API code.
The trade-off is overhead. MCP servers consume context tokens at initialisation, and poorly built servers can burn through your budget fast. Use them deliberately — only connect the tools your agents actually need.
The decision framework
| Question | Best fit |
|---|---|
| Is the value in how we do something? | Skill |
| Does the task involve chaining local tools? | CLI |
| Does the agent need authenticated access to a SaaS platform? | MCP |
| Is this for one person or the whole team? | Solo → CLI/Skill, Team → MCP |
| Does the user need to be technical? | Non-technical → MCP/Skill, Technical → CLI |
In practice, most marketing teams use all three in combination. A skill defines the process, a CLI handles the build step, and an MCP server pulls the data the agent needs. They’re layers, not choices.
What this means for Growth Method
Growth Method supports all three because no single approach covers every workflow.
- Skills encode your team’s growth methodology and processes
- MCPs connect your marketing stack so agents can access real data
- CLIs power the technical workflows that keep everything moving
But we’re thinking about this beyond just “supporting three interfaces.” The real opportunity is what happens when they work together.
Growth Method acts as both an MCP server and an MCP client. As a server, it exposes your campaign pipeline, strategy context, hypotheses, and performance learnings to any agent. A marketer working in Claude Code can ask “what’s our top priority campaign?” and the agent queries Growth Method for the answer. An automation running in n8n can pull the current pipeline every Monday morning. Whatever agent your team uses — today or two years from now — it can read from and write back to Growth Method through the same standard protocol.
As a client, Growth Method’s own agents connect to your existing stack — PostHog, Google Analytics, Search Console — to pull the data they need. A Campaign Analysis agent reads performance metrics from across your tools, compares them against the original hypothesis, and writes the learnings back. A Campaign Ideas agent searches your strategy context and past results to suggest what to try next. The agents don’t replace your marketing team. They connect the dots across your stack that nobody has time to connect manually.
Skills tie it all together. They encode how your team does growth — your analysis methodology, your ideation framework, your reporting structure — so every agent interaction reflects your way of working, not generic AI output.
The whole system is built on open standards. No proprietary connectors, no lock-in. If a tool speaks MCP, it works with Growth Method. If an agent supports skills, it can learn your methodology. That means you’re not betting on any single AI platform — you’re building on the infrastructure layer the entire industry is converging on.
The bottom line
MCPs, CLIs, and Skills aren’t competing standards — they’re complementary layers. Skills capture your team’s knowledge, CLIs give you speed, and MCPs give the whole team governed access to your stack. The teams that adopt AI most effectively will use all three where each fits best, rather than forcing every problem through a single interface.
Growth Method is the growth operating system for marketing teams focused on pipeline — not projects. Book a call to see how we can help accelerate your results.
“We are on-track to deliver a 43% increase in inbound leads this year. There is no doubt the adoption of Growth Method is the primary driver behind these results.”
Laura Perrott, Colt Technology Services