Why this workflow exists
If you publish to Instagram, the per-post overhead is brutal. A 5-slide carousel takes 30–60 minutes by hand — even when you already know what you want to say. Most of that time is mechanical: opening Figma, pasting brand colors, aligning text, exporting at the right size, naming files.
Claude can do all of that. What it needed was a connection to a tool that actually renders pixels. Pencil MCP is that connection. Once it’s wired in, your role shrinks to “tell Claude what to make” and “approve the previews”. Everything in between disappears.
The first time you run this workflow it takes ~30 minutes — most of that is one-time: installing Pencil MCP, telling Claude your voice, saving your brand kit. After that, it’s 15–20 minutes per carousel.
Install in 10 minutes
You need three things: Pencil MCP installed, optional Figma MCP for brand assets, optional Memory MCP for voice persistence.
1. Install Pencil MCP
# In your Claude Desktop / Cursor / Windsurf config:
claude mcp add pencil -- npx -y @pencil/mcp
Set PENCIL_API_KEY in the env block. Get it at pencil.so/account.
2. (Optional) Install Figma MCP
Same pattern. You only need this if you want Claude to read your existing Figma brand kit.
claude mcp add figma -- npx -y @figma/mcp
Set FIGMA_TOKEN (Personal Access Token from Figma → Settings → Account).
3. (Optional) Install Memory MCP
This is the official Anthropic memory server. It lets Claude remember your brand voice across sessions.
claude mcp add memory -- npx -y @modelcontextprotocol/server-memory
No env vars needed.
4. Load the skills
If you have the carousel-content-expert and visual-brief skills in your local skills directory (~/.claude/skills/), they’ll be loaded automatically. If not — they’re optional, the workflow works without them but the output is less polished.
The first run
Open a fresh Claude conversation. Paste step 1’s prompt, replacing the bracketed parts with your actual topic. Claude will start an outline. From here you’re approving and adjusting, not designing.
If the outline looks good, move to step 2. If not, push back: “make slide 3 stronger, it’s the same as slide 2.” This is editing, not generation.
By step 4 you have a folder of PNGs. Drag them into Instagram in order. Done.
What success looks like
You opened Claude with an idea. You closed Claude with a finished carousel folder. Total wall-clock time: 20–30 minutes. The brand looks like your brand. The voice sounds like your voice. You didn’t open Figma once.
Now do it again next week. The setup is one-time. The recipe is repeatable.
Going further
- Multiple platforms. Once the carousel is rendered, ask Claude to write a LinkedIn version of the same content (different shape — long-form post, no carousel). Use the
linkedin-posting-voiceskill if you have it. - Scheduled queue. If you publish multiple per week, ask Claude to add the finished carousel to your draft folder with a date-prefix filename. Hook into Buffer / Later / Postiz via their MCP if available.
- Voice consistency check. Ask Claude to score the new draft against your saved voice profile (in Memory MCP) and flag any drift before you publish.
Related
- Pencil MCP — the rendering engine
- Figma MCP — for brand asset access
- Content Creator stack — the full bundle this workflow lives in