Reference
A reference index of the plugins, skills, MCP connectors, and browser agents you can install on top of your AI platform. Built for operators getting beyond chat with Claude, ChatGPT, and similar.
The AI Vendor Landscape covers the platforms themselves. This page covers what to add on top: plugins (bundles of skills and tools), skills (single-purpose know-how), and MCP servers (connectors to your apps and data). Different shopping moment, different mental model. Pick a category, scan the cards, and do your own due diligence before installing anything.
No bolt-ons match your search. Try a different term, or submit one we should add.
Bundles of capability you install in one motion: a plugin can ship dozens of skills, prompts, and tools tuned for a domain. The highest-leverage install if you want a whole specialist team showing up at once.
23 specialist tools in a single install. The closest thing to dropping a full development team into your Claude in one move. Strong default if you don't know where to start.
A complete development methodology packaged as 14 skills. Opinionated about how to structure work, draft requirements, review code, and ship reliably. Most-starred plugin in the ecosystem.
OpenAI's official Codex plugin for Claude Code. Lets you run OpenAI's coding model side by side with Claude in the same session. Handy when you want a second opinion on a tricky diff.
Investment banking, equity research, private equity, and wealth management workflows in one bundle. Aimed at boutique shops and CFO offices that want senior-analyst leverage without a full data team.
Legal workflows across practice areas: contract review, clause comparison, redline drafting, and matter intake. Sized for solo practitioners and small firms that don't have a legal-ops team.
The catch-all plugin: 255+ skills across every platform. Native .docx, .pptx, .xlsx, and .pdf creation; image, audio, and video helpers; and a long tail of workflow primitives. The single highest-leverage install for knowledge workers.
40 marketing tools across the full growth-ops surface: campaign planning, persona briefs, channel-fit analysis, lifecycle copy, and reporting templates. Marketer hand-off without losing the strategy.
A content OS for posts, reels, and captions. Voice consistency across platforms, scheduling cadence helpers, and adaptation between LinkedIn, X, and short-form video. Pairs naturally with Granola for meeting-to-post pipelines.
Open-source marketplace for Claude Code: 471 plugins, 3,069 skills, and 347 agents in one registry, with the ccpi CLI package manager (search, install, update). One command gets you any community workflow instead of hunting through GitHub.
Single-purpose know-how you add to your AI: a skill is one well-described task, automatically invoked when the work calls for it. Lighter than a plugin, narrower than an MCP, and easy to mix and match.
Kills generic AI UI. The taste fixer for front-end work: opinionated about spacing, type, and component shape so your build looks designed instead of defaulted. Most-installed skill in the catalog.
Write HTML, render video. Agent-native. Lets Claude generate motion graphics and explainer-style video right out of HTML markup. The fastest path from rough idea to a usable B-roll clip.
Karpathy-style wiki built from your AI history. Turns past chats and projects into a queryable knowledge base so you stop re-asking the same setup questions.
Lets Claude query your research and playbooks the way NotebookLM does: grounded answers, source citations, and good behavior on long PDFs. Stays inside Claude so you don't need to swap tools.
Strips AI writing tells from any draft. Targets the specific cadence and word choices that make machine prose feel machine-written, and rewrites them in a more natural voice.
GEO-first SEO skill. Built for the AI era: optimizes pages for both classic search and the new generative-search surfaces (Perplexity, ChatGPT Search, Google AI Overviews).
The Vue and Vite core teams' own skills collection. Best-in-class for front-end work on those stacks because the maintainers wrote it. Use it any time you're building or refactoring a Vue app.
Cuts about 65% of tokens. Talks like caveman. Sounds like a joke, behaves like a cost lever: shorter prompts, shorter responses, less budget burn for routine calls.
Reusable skills for CRO, copy, SEO, analytics, launch, pricing, and growth workflows. Drops a marketing specialist into any agent session without the headcount.
Community design-intelligence skill that layers opinionated UI/UX guidance on top of Claude's output — more direction than the official frontend-design plugin when design quality is your bottleneck. Install via /plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill.
Karpathy's four coding rules as a drop-in CLAUDE.md: think before coding, simplicity first, surgical changes, goal-driven execution. 200k+ stars and the fastest-growing behavioral skill for coding agents.
Connectors that let your AI read and act on the rest of your stack. MCP (Model Context Protocol) is the open standard that finally makes "your AI can use my tools" a normal sentence instead of a custom build. Three flavors below: direct hooks into apps you already use, specialized tools that bring their own AI capability, and integration platforms that bundle thousands of services behind one endpoint.
Feeds every meeting note to Claude. Granola already records and summarizes your calls; the MCP lets Claude pull those notes into any later workflow without manual copy-paste.
Posts updates and reads channel history on your behalf. Workflows finally get real workplace context instead of you pasting threads in. The single most useful MCP for most teams.
Reads and writes your databases and docs. Claude can query Notion as a knowledge base, then update pages from a conversation. Best paired with operators who already live in Notion.
Flags which LinkedIn DMs actually need a reply. The first useful AI take on relationship inbox triage: skips the noise, surfaces the conversations that move work forward.
Token-efficient browser automation. A sandboxed browser the AI controls so it can fill forms and pull info from pages without a public API, without burning your context window on raw HTML.
Real-time web search inside Claude. Drops Perplexity's grounded, citation-first search behavior into any Claude session so research questions don't require swapping tools.
Cinematic video from a single prompt. The MCP exposes Higgsfield's model to Claude so you can describe a clip in conversation and get a usable rough cut without leaving the chat.
Lifelike text-to-speech and voice cloning. The MCP lets Claude turn written text into natural-sounding audio, produce voiceovers, and transcribe speech without leaving the conversation.
9,000+ apps behind one MCP endpoint. The lazy-but-correct way to bridge Claude to everything else in your stack: CRMs, billing tools, calendars, spreadsheets. Free tier covers 100 tasks per month (each MCP call counts as two), then paid plans scale from there.
250+ tools behind one MCP endpoint, built specifically for AI agents. Free tier covers 20,000 tool calls per month with no credit card; paid plans start at $29/month for 200,000 calls. Open-source under MIT, with both hosted and self-hosted options for teams that want their integrations to stay on their own infrastructure.
3,000+ APIs and 10,000+ prebuilt tools behind one MCP endpoint, with managed OAuth and per-user credential handling baked in. Free in development mode for testing; pricing kicks in when you move to production. Stronger on event-driven workflows than Zapier and more developer-leaning out of the gate.
Official hosted MCP server from X. 200+ API endpoints for tweeting, searching, DMing, and analytics. Released June 2026.
Dynamically fetches version-specific documentation for React, Next.js, Tailwind, and 100+ frameworks. 54k+ GitHub stars, 890K weekly npm downloads.
Official MCP server from GitHub with 9 configurable toolsets: repos, issues, PRs, Actions, security, discussions, deployments. PR/issue automation via natural language.
Official MCP server from Microsoft for browser automation. Stealth-mode browsing, screenshots, form fills, and cross-browser testing. 30k+ GitHub stars.
Builds a persistent knowledge graph of your codebase with tree-sitter across 158 languages - cuts structural-query token use by up to 99%. Single static C binary, runs fully local.
Official Figma MCP server. Gives Claude and other agents read access to design files — components, variables, and page structure — so design context flows straight into code generation without screenshots.
Official Linear MCP server. Agents can read, create, and update issues and projects in your Linear workspace — the system-of-record pick for teams that run product work there.
Official Stripe Agent Toolkit with MCP support. Agents can look up customers, inspect invoices and subscriptions, and answer billing questions against your live account with proper scoped keys.
Cloudflare's hosted MCP servers for Workers, D1, R2, and its other platform products. Day-zero support for the MCP 2026-07-28 spec (stateless, load-balancer friendly) and free to run your own remote MCP servers on Workers.
Tasks is one of the first official MCP extensions, contributed by AWS in the 2026-07-28 spec. Adds reliable, long-running agent operations — durable task state and resumable work — for production agent workloads.
Enterprise MCP deployment (July 8, 2026): Slackbot + Salesforce uses MCP to pull CRM data, build Tableau charts, and trigger DocuSign approvals directly inside Slack channels. Work stays where the conversation is — Salesforce's IT team says it saved thousands of custom coding hours annually.
Turns the web into agent-readable data: scrape, crawl, search, and extract structured JSON from any site. The default MCP for research agents that need clean data instead of a wall of HTML.
Cloud browser automation API: agents navigate real pages, click, fill forms, and read the results. Pairs with the open-source browser-use library that made LLM web navigation mainstream.
One MCP endpoint to hundreds of models. Route by task, compare outputs, and swap providers without changing your agent stack.
Official MCP server for the Hugging Face Hub: search models, datasets, and Spaces and pull metadata straight from your agent.