Tools

Claude Code plugins, skills and MCP servers I use

Not a link list. Each one is something I run in my own projects: what it does, how I use it, and what to watch out for.

Tools I use

gemini-cli (gemini-mcp-tool)

MCP

Lets Claude hand big reads (files, logs, folders) to Gemini and get back only a summary.

How I use it: A big help for saving tokens, and easy to set up.

Watch out: Gemini can be overloaded or retire a model name; when it errors twice, Claude has to do the job itself. On the free tier, Google may use prompts to improve its products.

claude mcp add -s user gemini-cli -- npx -y gemini-mcp-tool

graphify

Skill

Turns a folder into a knowledge graph (nodes, links, groups) plus an HTML view and a report.

How I use it: I ran it once on my Arttie design folder: 209 files became 615 nodes in 27 groups.

Watch out: That first run cost about 1.1 million tokens. If you're low on usage, don't run it on a whole project.

uv tool install graphifyy
graphify install --platform claude

Records what happens in each session and brings the relevant notes back in later sessions.

How I use it: When I wrote the guides on this site, the real details (how my GitHub profile README evolved, the Cloudflare build fix, a layout-shift score of 0.105) came from its history, not from me retyping them.

Watch out: It adds its own context to sessions and tool calls, which costs tokens. Its notes are summaries, so check anything important against the actual files. It also asks you to subscribe to its paid cmem.ai membership; you can skip that, and the plugin still works locally.

/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem@thedotmack

A "lazy senior dev" mode: the smallest change that fully solves the task, no extra features, and each reply ends with what was skipped.

How I use it: On for this site: changes stay small and every answer says what wasn't checked. It also shows the cost in the chat, which tells me how to write my next prompt to save tokens or give Claude a better-sized task.

Watch out: "Smallest change" isn't always what you want for design work. Switch it off, or to lite, when you want more polish.

/plugin marketplace add DietrichGebert/ponytail
/plugin install ponytail@ponytail

Runs ComfyUI from the terminal, so Claude can generate images on your own GPU. The MCP server adds tools for the rest: logs, restarts, freeing GPU memory.

How I use it: My comfyui-image skill picks the best model for the prompt (anime, photo or a quick draft), runs the workflow and shows the picture in the chat. It can also download a missing model for me. It saves tokens too: the image is made on my own GPU, so Claude only spends tokens on a few short commands, and I still get high-quality images.

Watch out: An exported workflow keeps the last prompt you typed, so clean it before sharing. Big images can run out of GPU memory.

uv tool install comfy-cli
uv tool install comfy-mcp

Design direction for building or reshaping UI, focused on avoiding the "AI-made" look.

How I use it: Mainly to review a design and find where it can be better. On this site, its list of AI-design tells (cream background, ALL-CAPS labels, an arrow on every link, fade-ins everywhere) drove the home page redesign.

Watch out: Strong opinions. It's a checklist to think with, not rules to follow blindly.

/plugin marketplace add anthropics/claude-code
/plugin install frontend-design@claude-code-plugins

A searchable library of UI/UX rules, palettes, font pairings and page patterns, run through a local Python script.

How I use it: Mainly to review a design and find where it can be better. On this site it caught the usability gaps: touch targets under 44px, contrast, page structure.

Watch out: Searches need the right keywords; my first search returned nothing and needed a retry. Needs Python installed.

/plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill
/plugin install ui-ux-pro-max@ui-ux-pro-max-skill

Running in the background

I never call these myself. They're installed so Claude picks them up on its own when a task fits, a quiet upgrade to how useful it is. The catch: you rarely see when they kick in, so it's hard to judge how much they help. And every installed skill adds its description to each session, so remove the ones that never seem to fire.

claude-code-setup

Plugin

Looks at a codebase and recommends Claude Code automations: hooks, subagents, skills, plugins, MCP servers.

Watch out: Its skill description sits in every session; remove it if it never seems to fire.

/plugin install claude-code-setup@claude-plugins-official

Security review of the code Claude writes: warnings on each edit, an AI review of the diff when Claude finishes, and a commit reviewer. Covers injection, XSS, SSRF, hardcoded secrets and 25+ other issues.

Watch out: Mostly silent until it finds something, so it's hard to judge how much it catches.

/plugin install security-guidance@claude-plugins-official

design-taste-frontend (taste skills)

Skills

Art direction for landing pages and portfolios, with a library of researched directions and a screenshot look-and-fix loop.

Watch out: Several versions get installed side by side (v1, v2, v2-experimental), and they're easy to mix up.

npx skills add Leonxlnx/taste-skill