Six non-coding things you can do with Claude Code

Claude Code stops being a coding tool the moment you wire it up to the services you actually use. The feature that unlocks this is called MCP — Model Context Protocol — and most people don’t understand it yet.

An MCP is a wire. It connects Claude to one specific service. After it’s installed, you stop typing things like “here’s my data, please analyze it” and start typing things like “look at my last 90 days in Google Analytics and tell me what’s converting.”

Here are six workflows that are very much not coding. Each one runs on one or two MCPs. None of them require you to write code.

1. SEO briefings from your own analytics

MCPs needed: Ahrefs, GA4 (Google Analytics), Google Search Console.

You have data flowing into three different dashboards that you almost never log into. Claude Code can.

The prompt looks like this:

Pull the last 30 days from GA4. Cross-reference with Search Console queries. Find pages where impressions are climbing but conversion is flat. Give me a three-bullet brief on what to try next.

The output is a briefing, not a dashboard. It tells you the page, the trend, the suspected gap, and a specific change to test. Three minutes of reading replaces an hour of staring at colored bars.

Setup:

claude mcp add ahrefs --transport http <ahrefs-mcp-endpoint>
claude mcp add ga4 --transport http <ga4-mcp-endpoint>
claude mcp add gsc --transport http <gsc-mcp-endpoint>

Then ask in plain English.

2. Niche video research

MCPs needed: Browserbase, Apify.

If you make video, write a newsletter, or just want to know what topics are working in a corner of the internet, this is the workflow.

Find videos in [niche] that crossed 100k views in the last 90 days. Sort by recency. Give me title, channel, and a one-line read on why it worked.

Browserbase opens browser sessions Claude can drive. Apify gives it scraping infrastructure. Together they can sweep the platform and bring back a hit list without you opening YouTube.

This is also useful as a “what would I make a video about” tool. You feed it your niche, you get back what’s already pulling views, you decide what angle is missing.

3. PRDs that auto-populate Notion and Linear

MCPs needed: Obsidian, Notion, Linear.

Writing a product requirements doc is a chore. So is creating the matching wiki page in Notion. So is filing the implementation tickets in Linear. That’s three tools and ninety minutes for something nobody enjoys.

With all three MCPs installed:

Read the feature brief I just wrote in feature-x.md. Write a full PRD. Publish it as a Notion page in the Engineering workspace. Create Linear tickets for each implementation chunk, assigned to the relevant person.

One prompt. Three tools updated. The doc, the wiki, and the tickets in sync because they came from the same source.

This compounds. The fourth time you write a PRD, the prompt becomes a slash command. Then it’s /prd <filename> and you stop thinking about it.

4. document editing as a thinking partner

MCPs needed: none. Just Claude Code in the folder where your draft lives.

You don’t always need MCPs. Sometimes you just need a teammate who can read your full draft and push back.

Open Claude Code in the folder where your essay or memo or proposal lives. Ask:

Read q3-strategy.md. Tell me what argument I’m avoiding. Tell me which paragraph is the weakest. Suggest three structural changes that would make it shorter.

This is not the same as asking ChatGPT to “improve” something. You’re not asking for a rewrite. You’re asking for diagnostic feedback on a real file, in place, that you then edit yourself.

The file lives on your laptop. The edits live on your laptop. Nothing gets pasted anywhere.

5. YouTube transcripts → notes

MCPs needed: Apple Notes MCP, plus the yt-dlp or youtube-transcript CLI.

You watched a long video. Three things in it were useful. You will never find them again unless you take notes. But you’re not going to take notes.

The workflow:

Pull the transcript from [youtube URL]. Summarize the three sharpest points with timestamps. Save the note in Apple Notes under “Watched - YYYY-MM.”

The transcript gets pulled by the CLI. Claude reads it. The summary lands in your Notes app, organized by month, searchable. The transcript is also archived locally so you can re-query it later.

6. Domain brainstorming with live availability

MCPs needed: a domain availability MCP (several open-source ones exist).

You’re naming a product or a side project. You generate 40 names. You then spend an hour checking which ones are available. By the time you finish, your favorites are gone.

Brainstorm 50 domain names for [project description]. Check .com, .ai, .io availability live as you go. Mark the ones I can buy right now with a green flag. Tell me the trademark conflicts on any flagged green.

You get back a list with availability flags and conflict notes. You skip the soul-crushing part of the naming process.

The meta-pattern

Re-read those six and notice what changed. The prompts are not clever. The setup is not complex. The unlock is access.

You are not writing better prompts. You are giving Claude the keys to a service you already pay for and asking in plain English.

The mental shift: stop thinking about Claude Code as “an AI.” Start thinking about it as a teammate with read-write access to whatever you wire it up to. The MCPs are the wires. The prompts are the briefings.

How to start

1. Pick one annoying recurring task tied to one service you already use.
2. Install that service’s MCP. Most are a single claude mcp add command.
3. Ask in plain English. Describe the outcome, not the steps.
4. If the prompt works twice, save it as a slash command.

That’s it. That’s the whole on-ramp.

Further reading

• How MCPs work in Claude Code
• What a Claude Code skill is
• Deepak Ness’s original list of non-coding tasks