Awesome Agent Plugins

English | Português

One catalog of MCP servers, Agent Skills and native plugins for AI coding agents — with the exact install command for your client.

Most lists are tied to one client. Here every entry says which agents it works with, and install instructions are generated per client from a single canonical spec, so they stay consistent. Near-duplicate plugins are folded into one entry that points to the best option and lists what each alternative adds.

Supported clients

Client	MCP	Skills	Native plugins	Project instructions	Install commands
Claude Code	✅	✅	✅ 3	CLAUDE.md	✅ tested end-to-end
Codex CLI	✅	✅	—	AGENTS.md	✅ tested end-to-end
Gemini CLI	✅	✅	✅ 3	GEMINI.md	✅ tested end-to-end
Cursor	✅	✅	—	AGENTS.md / .cursor/rules	from official docs
VS Code (Copilot)	✅	✅	—	.github/copilot-instructions.md	from official docs
OpenCode	✅	✅	—	AGENTS.md	from official docs
Cline	✅	✅	—	.clinerules	from official docs
Windsurf	✅	✅	—	AGENTS.md	from official docs
Zed	✅	✅	—	AGENTS.md	from official docs
Goose	✅	✅	—	.goosehints	from official docs
Kiro	✅	✅	—	.kiro/steering/	from official docs
Roo Code	✅	✅	—	.roo/rules/ + AGENTS.md	from official docs
DeepSeek Harness	✅	✅	✅ 221	AGENTS.md	from official docs

“Tested end-to-end” means npm run e2e runs the generated MCP and skill install commands against the real CLI in a throwaway directory. Other clients follow their official documentation (sources).

Quick start (CLI)

Search the catalog and get (or run) the install command for your agent:

npx github:LMPrado-DZ23/awesome-agent-plugins search browser
npx github:LMPrado-DZ23/awesome-agent-plugins show playwright-mcp
npx github:LMPrado-DZ23/awesome-agent-plugins install playwright-mcp --client codex        # print the command/config
npx github:LMPrado-DZ23/awesome-agent-plugins install anthropic-skills --client claude-code --run

Featured

Name	Type	What it does	Works with
Superpowers ★289k	Agent Skills	Skills library and development methodology: brainstorming, planning, TDD, systematic debugging, code review and subagent-driven execution.	All clients
Anthropic Skills ★177k	Agent Skills	Reference Agent Skills from Anthropic: PDF, DOCX, XLSX and PPTX handling, skill-creator, MCP builder, frontend design, web-app testing and more.	All clients
Context7 ★62k	MCP server	Up-to-date, version-specific library documentation and code examples pulled into the agent context on demand. Local alternative: npx -y @upstash/context7-mcp.	All clients
Chrome DevTools MCP ★52k	MCP server	Gives agents Chrome DevTools: drive a live Chrome, inspect network and console, record performance traces and debug pages.	All clients
Playwright MCP ★37k	MCP server	Browser automation for agents via Playwright accessibility snapshots: navigate, click, fill forms, take screenshots, without needing a vision model.	All clients
Claude Code official plugin directory ★36k	Plugin marketplace	Anthropic-managed directory of Claude Code plugins (commands, agents, skills, hooks, MCP servers) installable with /plugin.	Claude Code only
GitHub MCP Server ★33k	MCP server	GitHub's official server: repositories, issues, pull requests, Actions, code scanning and more. Remote endpoint shown; a Docker image (ghcr.io/github/github-mcp-server) runs it locally.	All clients
skills (npx skills) ★32k	Developer tool	CLI that installs, updates, lists and removes Agent Skills from any Git source into 75+ agents (Claude Code, Codex, Cursor, Gemini CLI, OpenCode…). Used by this catalog for skill installs.	—
Serena ★30k	MCP server	Semantic code retrieval and editing through language servers: find symbols and references, and edit at symbol level instead of by text.	All clients
Agent Skills ★25k	Standard / spec	Specification for Agent Skills: a folder with a SKILL.md (name, description, instructions) plus optional scripts and resources that an agent loads on demand.	—
AGENTS.md ★24k	Standard / spec	Open Markdown format for project instructions to coding agents (setup, build, test, conventions). Read by Codex, Cursor, OpenCode, Zed, Windsurf, DeepSeek Harness and others.	—
Model Context Protocol ★9.3k	Standard / spec	Specification and documentation for MCP, the open protocol every client in this list uses to connect agents to tools, data and prompts.	—
MCP Registry ★7.3k	Curated list / registry	Official community-driven registry service for MCP servers, with a public API for discovering published servers and their install metadata.	—
dsh-market ★4.1k	Plugin marketplace	Plugin market inside DeepSeek Harness Settings: search, one-click install/upgrade and theme switching for every plugin in awesome-dsh-plugin.	DeepSeek Harness only
DZ23 Subagents Universal MCP ★0	MCP server	Self-hosted MCP router that delegates text/code tasks to many models with automatic failover, parallel role-based subagents (architect, backend, security, QA…), consensus and mission memory shared across harnesses. Node.js 22+, no runtime dependencies.	All clients
Enterprise Product Builder	Agent Skills	Skill for building web products to mature-company standards: discovery and metrics, ADRs, API/data contracts, design system, OWASP ASVS security, LGPD/GDPR privacy, WCAG 2.2 AA, i18n, SLOs, CI/CD with rollback and release review, plus templates.	All clients
Super Autonomous Mission Loop	Agent Skills	Skill for long engineering missions that run until objective completion criteria: bounded autonomy, on-disk checkpoints, recovery after crash or context compaction, anti-loop rules, Git discipline, three independent audits and verified release.	All clients

Collections

Ready-made kits for common setups — install a whole kit on one client with a single command.

• Autonomous engineering (DZ23 stack) — Multi-model subagents with shared mission memory, long-mission and product-quality skills, responsible AI-origin review, and GitHub access. (aap collection autonomous-engineering --client <id>)
• DeepSeek Harness starter kit — The plugin market, the most-used native dsh plugins (sidebar, context dashboard, agent teams, memory, auto-review) and portable MCP servers and skills that also work in dsh. (aap collection deepseek-harness-starter --client <id>)
• Full-stack web development — Up-to-date docs, a real browser for testing, semantic code navigation, GitHub and error tracking. (aap collection fullstack-dev --client <id>)
• Building your own MCP servers and skills — The specs, an inspector to debug servers, a Python framework, and the CLI to publish/install skills. (aap collection mcp-builder --client <id>)
• Research and documents — Web search and scraping, converting files to Markdown, and skills for PDF/Office documents. (aap collection research-docs --client <id>)

Contents

• Standards & Specs — 3 entries
• Registries, Lists & Plugin Managers — 18 entries + 17 alternatives folded in
• Skills & Instructions — 46 entries + 49 alternatives folded in
• Code Intelligence — 10 entries
• Browser & Web — 17 entries + 22 alternatives folded in
• Docs & Knowledge — 29 entries + 12 alternatives folded in
• Memory — 25 entries + 18 alternatives folded in
• Tools & Capabilities — 61 entries + 33 alternatives folded in
• Development & Runtime — 78 entries + 24 alternatives folded in
• Git & Code Review — 16 entries + 11 alternatives folded in
• Cloud, Data & Infra — 40 entries + 1 alternatives folded in
• Workflow & Agents — 46 entries + 24 alternatives folded in
• Integrations & Notifications — 61 entries + 13 alternatives folded in
• Security & Permissions — 33 entries + 28 alternatives folded in
• Models & Providers — 13 entries + 18 alternatives folded in
• Usage & Cost — 8 entries + 8 alternatives folded in
• Sessions & Messages — 10 entries + 7 alternatives folded in
• Vision & Voice — 31 entries + 21 alternatives folded in
• Remote & Mobile — 11 entries + 19 alternatives folded in
• UI Enhancements — 20 entries + 21 alternatives folded in
• Themes & Appearance — 8 entries + 6 alternatives folded in
• Just for Fun — 5 entries + 4 alternatives folded in

Where entries come from

Every import is reproducible (scripts/import/) and every entry passes the same validator. Importers only produce candidates; a review step groups duplicates and writes the descriptions.

Source	Entries	How it is selected
Hand-curated	44	Official servers, specs and skill packs checked against their READMEs.
MCP Registry	198	Servers from the official registry with ≥200★, whose registry namespace matches the GitHub owner (anti-impersonation); install spec taken from the registry package data.
GitHub discovery	127	Topic/keyword search, classified by files actually present (SKILL.md, plugin marketplace, Gemini extension manifest).
awesome-dsh-plugin	220	Top plugins by stars and npm downloads; near-duplicates folded into alternatives.

DeepSeek Harness plugins

Imported from awesome-dsh-plugin: the 220 strongest of ~3,900 plugins (ranked by GitHub stars and npm downloads, archived/missing repos dropped), with 193 similar plugins folded in as alternatives. Install with dsh plugin add; browse the original list for everything else.

How entries are chosen

• It works: installs as described on at least one client, and the description matches the code.
• It is maintained: archived or deleted repos are flagged by a weekly health check and removed.
• No duplicates: when several projects solve the same problem, the best one gets the entry and the others appear as alternatives with what they add.
• No secrets: configs only contain placeholders such as <GITHUB_PAT>.
• Listing is not a security review — third-party code runs with your permissions. Read the source first.

Contributing

Anyone can contribute — no permission needed. Add one YAML file under data/entries/ and open a PR (READMEs and catalog pages are generated; see CONTRIBUTING.md), pick a good first issue, or ask and suggest in Discussions.

License

Catalog data: CC0-1.0. Code (CLI and scripts) and the skills/instructions hosted in skills/ and instructions/: MIT. Each listed project keeps its own license.