Section 1
What is an MCP server?
MCP stands for Model Context Protocol. It's a feature Anthropic shipped that lets Claude talk to other apps and services — without you copy-pasting anything between tabs.
Think of an MCP as a plug-in. Each one gives Claude a new capability: search the live web, generate images, scrape a website, read your calendar, post to Slack. Install three or four and Claude stops being a thing you talk to and starts being a machine that runs work for you.
This guide walks you through the three MCPs that move the needle the most for creators and operators: Perplexity (live research), Nano Banana (image generation), and Apify (web scraping).
Each section covers the setup for both Claude Code (CLI) and Claude Desktop, plus three copy-paste prompts you can run the second you're connected.
About the prompts. Every prompt in this guide uses [SQUARE_BRACKET] placeholders. Before you run a prompt, replace every bracketed word with your own input — your topic, your competitor, your city, your file path.
MCP #1
Perplexity — Live Research
Claude has a knowledge cutoff. Perplexity fixes that. Once installed, Claude can search the live web, cite sources, and pull real-time data into whatever you're working on. Fact-checking, competitor research, trend reports — this is the one you run the most.
Step 1 — Get your API key
Go to console.perplexity.ai → API Keys → Generate. Copy the key. Pay-as-you-go pricing, starts at $5 per 1,000 searches. Perplexity Pro subscribers get $5/mo in API credits included.
Step 2a — Install in Claude Code (CLI)
$ claude mcp add perplexity --env PERPLEXITY_API_KEY="your_key_here" -- npx -y @perplexity-ai/mcp-server Step 2b — Install in Claude Desktop
Open claude_desktop_config.json (Settings → Developer → Edit Config) and add this to mcpServers:
"perplexity": {
"command": "npx",
"args": ["-y", "@perplexity-ai/mcp-server"],
"env": { "PERPLEXITY_API_KEY": "your_key_here" }
}Restart Claude. You should see perplexity_search, perplexity_ask, perplexity_research, and perplexity_reason appear in the tools list.
3 prompts to run first
Every [PLACEHOLDER] gets replaced with your own input before you run it.
Prompt 1 — Weekly trend report
Using perplexity_research, pull together a trend report on [TOPIC/NICHE] covering the last 30 days. I want:
1. The 5 most-discussed stories, tools, or shifts (with publish dates)
2. Who's talking about them (creators, publications, companies)
3. What the consensus take is vs. the contrarian take
4. 3 content angles I could post this week that would feel timely
Cite every source. Keep it tight — no filler.Prompt 2 — Fact-check before publish
Here's the draft I'm about to publish:
[PASTE YOUR DRAFT HERE]
Using perplexity_ask, verify every statistic, feature claim, pricing mention, and name in this draft. For each one:
- Confirm it's still accurate as of today
- Flag anything outdated, wrong, or unverifiable
- Give me the corrected number or claim with a cited source
Return a bulleted list of every fix I need to make before I hit publish.Prompt 3 — Competitor pulse
Using perplexity_research, give me a competitive intelligence report on [COMPETITOR NAME or COMPETITOR URL].
Cover:
1. What they've launched or announced in the last 90 days
2. Current pricing and offer structure
3. The angle they're using in their marketing (what hook, what promise)
4. Weaknesses or gaps I could differentiate against
5. Who their ICP actually is based on how they talk publicly
End with 3 specific ways I could position [MY BRAND/OFFER] against them.MCP #2
Nano Banana — Image Generation
Nano Banana is Google Gemini's image model, wrapped as an MCP so Claude can generate and edit images directly in-chat. Thumbnails, carousel slides, product mockups, ad creatives — all without leaving the conversation. Kills the ChatGPT → Midjourney → Canva loop in one install.
Step 1 — Get your API key
Go to aistudio.google.com/app/apikey, sign in with a Google account, click Create API key, and copy it. Free tier is generous enough for daily creator use. No credit card required to start.
Step 2a — Install in Claude Code (CLI)
We're using the open-source ConechoAI/Nano-Banana-MCP server:
$ claude mcp add nano-banana --env GEMINI_API_KEY="your_key_here" -- npx -y nano-banana-mcp Step 2b — Install in Claude Desktop
Add this block to claude_desktop_config.json:
"nano-banana": {
"command": "npx",
"args": ["-y", "nano-banana-mcp"],
"env": { "GEMINI_API_KEY": "your_key_here" }
}Restart Claude. You'll now have generate_image, edit_image, and continue_editing.
3 prompts to run first
Replace every [PLACEHOLDER] with your own input.
Prompt 1 — On-brand thumbnail
Use generate_image to create a thumbnail for a short-form video titled "[VIDEO TITLE]".
Style: [DESCRIBE YOUR BRAND — e.g. clean, modern, off-white background, warm tan accent color #C4856C, editorial serif headline, minimal]
Subject: [WHAT SHOULD BE IN THE FRAME — e.g. a laptop showing a terminal window, a phone with a chat interface, a stylized icon of the tool]
Text overlay: "[3-5 WORD HEADLINE]"
Aspect ratio: 9:16 vertical
Make it feel scroll-stopping but not clickbait. No generic AI gradient backgrounds.Prompt 2 — Carousel slide set
Using generate_image, create 5 Instagram carousel slides (1080x1350) for a post about [TOPIC].
Slide 1: Hook — "[HOOK LINE]" as the only text, big serif headline, clean off-white background
Slides 2-4: One key idea per slide, short headline + one supporting sentence, consistent layout
Slide 5: Soft CTA — "[CTA TEXT]" with my handle [@YOUR_HANDLE] in a subtle corner
Use continue_editing between slides so the visual style stays identical across all 5. Single accent color: [#HEX_CODE]. No stock-photo look.Prompt 3 — Product mockup from a reference
I'm attaching [PATH TO REFERENCE IMAGE] as the reference for my product/brand.
Use edit_image to create a hero banner that shows [PRODUCT OR CONCEPT] as if it were photographed for a premium landing page.
- Keep the exact color palette and texture from the reference
- Subject centered, soft natural light, shallow depth of field
- Leave the top-left third empty for headline text
- Output 1920x1080
Make it feel like a real product shot, not a rendered 3D mockup.MCP #3
Apify — Web Scraping
Apify is a library of thousands of pre-built scrapers (they call them "Actors"). With the MCP connected, Claude can pull every short-form video from any creator in your space, scrape a Google Maps list for local leads, or extract data from almost any public website — all with plain English instructions. This is the MCP that powers a content engine.
Step 1 — Get your Apify account
Go to console.apify.com/sign-up and create an account. No credit card required. Free plan includes $5 in platform credits every month — enough for hundreds of small scrapes. The Apify MCP uses OAuth, so you won't need to manually grab a token for Claude Desktop.
Step 2a — Install in Claude Code (CLI)
The Apify MCP is a hosted HTTP server. Install with:
$ claude mcp add --transport http apify https://mcp.apify.com First time you use it, a browser window opens and asks you to authorize Claude against your Apify account. One click and you're done.
Step 2b — Install in Claude Desktop
Add this block to claude_desktop_config.json:
"apify": {
"url": "https://mcp.apify.com"
}Restart Claude. First run triggers the OAuth flow in your browser. After that, Claude has access to the entire Apify Actor library — TikTok, Instagram, Google Maps, Amazon, LinkedIn, and thousands more.
3 prompts to run first
Replace every [PLACEHOLDER] with your own input.
Prompt 1 — Content engine: pull every short-form from a competitor
Using the Apify MCP, run the "clockworks/tiktok-profile-scraper" Actor against [@COMPETITOR_TIKTOK_HANDLE]. Pull their last 50 videos.
Return the results as a table with these columns:
- Video URL
- Caption / first line of the hook
- Views, likes, comments, shares
- Posted date
- Hashtags used
Then sort by views descending, pull the top 10, and tell me:
1. What hook patterns show up in the top performers
2. What topics or angles are hitting vs. missing
3. 5 video ideas I could make tomorrow based on what's clearly working for them
This is the raw fuel for a content engine — treat it like that.Prompt 2 — Local lead list from Google Maps
Using the Apify MCP, run the "compass/google-maps-scraper" Actor. Search for "[BUSINESS TYPE]" in "[CITY, STATE]" and return the top 100 results.
For each result, give me:
- Business name
- Website URL
- Phone number
- Email (if visible)
- Star rating + review count
- Address
Save the full list as a CSV at [~/Desktop/leads-[CITY].csv].
Then flag the top 20 most promising leads based on:
- Has a website (so they have some digital presence)
- 4.0+ rating with at least 20 reviews (so they're real and decent)
- No obvious signs of being part of a big chain
These are my outreach targets.Prompt 3 — Instagram competitor swipe file
Using the Apify MCP, run the "apify/instagram-scraper" Actor against these handles: [@HANDLE_1], [@HANDLE_2], [@HANDLE_3].
For each account, pull their last 30 Reels and return:
- Reel URL
- Caption (full)
- View count, likes, comments
- Posted date
- First frame / thumbnail description
Then build me a swipe file grouped by hook pattern. Cluster the top 20 reels across all three accounts into 3-5 hook archetypes (e.g. "contrarian take", "stat reveal", "before/after"). For each archetype give me:
1. The pattern in one sentence
2. 2-3 example hooks from the data
3. A fresh hook I could use for my next reel in that patternQuick Reference
All three, one page
Perplexity — Live research
$ claude mcp add perplexity --env PERPLEXITY_API_KEY="your_key" -- npx -y @perplexity-ai/mcp-serverGet key: console.perplexity.ai → API Keys
Nano Banana — Image generation
$ claude mcp add nano-banana --env GEMINI_API_KEY="your_key" -- npx -y nano-banana-mcpGet key: aistudio.google.com/app/apikey
Apify — Web scraping
$ claude mcp add --transport http apify https://mcp.apify.comAccount: console.apify.com/sign-up (free, $5/mo credits)
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.