Skip to content
BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
.github
.github
 
 
CONTRIBUTING.md
CONTRIBUTING.md
 
 
LICENSE
LICENSE
 
 
LICENSE-CONTENT.md
LICENSE-CONTENT.md
 
 
README.md
README.md
 
 
banner.png
banner.png
 
 

Repository files navigation

Awesome Agent Plugins

Awesome Plugin count Last commit Spec version License

A curated, verified directory for Agent Plugins — the open, vendor-neutral standard (published 2026-08-06 by Amazon, Anysphere/Cursor, GitHub, Microsoft, OpenAI and Vercel, with Google as a core maintainer) that bundles Agent Skills + MCP servers into one portable, drop-in folder. Every entry below is checked against a live plugin.json — no link dumps, no dead repos.

Contents

What is an Agent Plugin?

An Agent Plugin is a directory with a plugin.json manifest at its root. It's the packaging format the Agent Plugins Working Group (Amazon, Cursor, GitHub, Microsoft, OpenAI, Vercel) published on 2026-08-06 to give Agent Skills and MCP server configs one portable, install-once shape that every compliant client can read — instead of writing a separate integration for each client.

plugin-name/
├── plugin.json               # REQUIRED — $schema + name. Optional: version, description,
│                              # license, keywords[], author, homepage, repository, extensions
├── skills/
│   └── <skill-name>/
│       ├── SKILL.md           # one skill per immediate child dir — NOT searched recursively
│       ├── scripts/
│       └── references/
├── mcp.json                   # OPTIONAL — mcpServers{}: stdio | streamable-http | sse
└── com.vendor.client/         # OPTIONAL — namespaced client-specific extensions

Minimal valid plugin.json:

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "my-first-plugin"
}

What the spec deliberately does not cover — and this is the part worth being precise about, because it's what most secondary write-ups blur: Agent Plugins is a packaging format only. It defines the manifest, the fixed component locations, and two runtime variables (${PLUGIN_ROOT}, the package root; ${PLUGIN_DATA}, a persistent writable directory) that clients expand in args, env values and cwd — never in command, URLs, or header keys. It has no opinion on distribution, marketplaces, permissions, or how a client installs a plugin — each client (VS Code, Cursor, ChatGPT/Codex, GitHub Copilot, Kiro, and Google's tooling as core maintainer) owns that layer itself. It also enforces real security boundaries: every path a plugin ships must resolve inside the plugin root (symlink escapes are rejected), command in mcp.json is a single executable token that is never shell-interpolated, and there's no portable field for embedding credentials.

Quickstart: build your first plugin in 5 minutes

mkdir -p hello-plugin/skills/say-hello
cd hello-plugin

cat > plugin.json <<'EOF'
{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "hello-plugin",
  "version": "0.1.0",
  "description": "A minimal Agent Plugin — one skill, no MCP server.",
  "license": "MIT"
}
EOF

cat > skills/say-hello/SKILL.md <<'EOF'
---
name: say-hello
description: Greet the user by name and explain what this plugin does.
---

# Say Hello
1. Ask for the user's name if not already known.
2. Reply with a friendly greeting.
3. Mention that this plugin came from a `SKILL.md` file with zero extra config.
EOF

# Validate against the canonical schema before you ship it (see Tools below)
npx -y ajv-cli@5 validate --spec=draft2020 \
  -s https://agent-plugins.org/schemas/1.0.0/plugin.schema.json \
  -d plugin.json

Point a compliant client at the hello-plugin/ folder (in VS Code: Chat: Install Plugin From Source) and you have a working plugin — no mcp.json required, since it's optional. Add one when you actually need an MCP server; see upstash/context7 below for a real mcp.json in production.

Client support matrix

Verified against each vendor's own docs/announcement as of 2026-08-12. ✅ confirmed · ❓ not yet confirmed in public docs — please open a PR with a source if you can confirm it.

Client Skills MCP servers Client extensions (com.vendor.*) Install path
VS Code ✅ ✅ ❌ ignored today (docs) @agentPlugins in Extensions view · Chat: Install Plugin From Source · auto-appears after install via Copilot CLI. Requires chat.plugins.enabled.
GitHub Copilot (VS Code, CLI, app) ✅ ✅ ❓ Shared install path with VS Code above; also via Copilot CLI.
Codex / Codex CLI ✅ (discovers + reads plugin skills) ✅ (opt-in MCP 2026-07-28 protocol, paginated discovery) ❓ Codex CLI workspace plugin install; marketplace details unconfirmed.
ChatGPT ✅ (plugins can bundle skills) ❓ ❓ Install UX not independently confirmed at time of writing — see Codex Knowledge Base writeup.
Cursor ✅ ✅ ❓ Listed as a launch client by Vercel's announcement; specific install flow not independently confirmed — PRs welcome.
Kiro ✅ ✅ ❓ Via "Kiro Powers" installable packages (AWS Open Source Blog).

Google joined as a core maintainer on launch day (not a launch client) and is integrating support starting with Agents CLI and the Data Agent Kit.

If you run one of the ❓ cells day to day, open a PR — a one-line source link is all we need to flip it to ✅.

⭐ Plugin of the Week

context7 by Upstash — pulls version-specific library documentation straight into your agent's context instead of letting it guess API shapes from stale training data. It ships a real mcp.json (one streamable-http server) plus a skills/ entry, and it's one of the first production plugins verified against the 1.0.0 schema outside the launch clients' own repos. Install: point your client at the plugins/agent-plugins/context7 folder in the repo above.

Rotates weekly. Nominate an entry by opening an issue with the pick-of-the-week label.

The catalog

Format: - [name](repo-url) by [author](author-url) — one-line description. **[tag]** Tags: production (used in the wild) · beta · experimental · reference (spec/example, not meant to run standalone)

Official & Reference

  • agent-plugins-spec by Agent Plugins Working Group — the canonical v1.0.0 spec text, JSON schemas, and governance docs. [reference]
  • awesome-copilot by GitHub — GitHub's own collection of 94 compliant plugins spanning languages, cloud platforms and workflows; still the single largest source of real plugins, though no longer the majority of this catalog. [production]

Dev & Coding

Data & APIs

  • dak by Google — data-engineering skills for Google Cloud: pipeline architecture, dbt transforms, Spark/BigQuery SQL notebooks. [production]
  • database-data-management by Awesome Copilot Community — PostgreSQL/SQL Server administration, optimization and data-management guidance. [production]
  • hindsight by Vectorize — long-term agent memory (retain/recall/reflect) exposed as portable MCP tools. [production]
  • neon-postgres by Neon — manage a Neon serverless Postgres backend: branching, object storage, functions, AI Gateway. [production]
  • qdrant by Qdrant — stores and retrieves semantic memories via vector search. [production]
  • zillapi by Zillapi — live Zillow property data: Zestimates, full records, comps and listing search on 160M+ U.S. homes. [production]

Search & Research

  • context7 by Upstash — version-specific library documentation pulled straight into LLM context. [production]
  • exa by Exa — real-time web search, code search and web crawling with configurable tool selection. [production]
  • scientific-agent-skills by K-Dense Inc. — 158 ready-to-use scientific and research skills across biology, chemistry, medicine and drug discovery. [production]
  • stayingapi by StayingAPI — search, availability, pricing and cross-OTA price comparison across Airbnb, Booking.com, Vrbo and Google Hotels. [production]

Browser & Automation

Productivity

DevOps

Content & Media

  • skill-image-gen by adamd9 — AI image generation (OpenAI gpt-image-2, Google Gemini) from inside your coding workflow, BYO API key. [production]
  • transcriptapi by TranscriptAPI — YouTube transcripts, video and channel search, playlist extraction, no API key to manage. [production]

Security

Finance

  • dodopayments by Dodo Payments — official Dodo Payments plugin: 17 integration skills (checkout, subscriptions, billing, refunds) plus API and docs MCP servers. [production]
  • open-market-data by anotb — read-only stock, SEC, crypto and macroeconomic data with normalized provenance. [production]

Skills & MCP servers ready to be packaged

The Agent Plugins ecosystem is six days old at the time of writing, so most of the world's best Agent Skills and MCP servers aren't plugins yet — they just need a plugin.json dropped on top. This section is not the catalog above: nothing here has shipped a compliant manifest. It's a punch list of high-quality, real, maintained skills/MCP servers that would make excellent plugins, listed here so day-one readers still get something useful instead of an empty repo. Once one of these ships a plugin.json, it graduates to the catalog above (via PR — see Contributing).

TranscriptAPI, Zillapi and StayingAPI graduated out of this section this update — they now ship compliant plugins and are listed in the catalog above instead.

Tools

  • ajv-cli against the canonical schemas — the validation one-liner multiple plugin authors in the catalog above ship in their own CI: 
    npx -y ajv-cli@5 validate --spec=draft2020 \
  -s https://agent-plugins.org/schemas/1.0.0/plugin.schema.json -d plugin.json
npx -y ajv-cli@5 validate --spec=draft2020 \
  -s https://agent-plugins.org/schemas/1.0.0/mcp.schema.json -d mcp.json
  • eng/create-plugin.mjs by GitHub — the scaffolder GitHub's own contributors use to generate new plugins inside awesome-copilot; a good reference for writing your own.

This section is intentionally short. If you maintain a standalone Agent Plugins validator, linter, or scaffolder, open a PR — it's an empty niche and first-mover here is real.

Spec & resources

🛡️ Security notice

This is a curated list, not an audit. Entries here are verified to have a real, schema-valid plugin.json and a resolving repository at the time they were added — that is a packaging check, not a security review. The spec itself enforces path containment and forbids shell-interpolated commands and embedded credentials, but a plugin can still request an MCP server that talks to a service you don't trust, or a skill that gives bad instructions. Read a plugin before you install it, the same as you would a package or a browser extension. Report a plugin that misbehaves via an issue and we'll pull it pending review.

🤝 Contributing

PRs are very welcome — see CONTRIBUTING.md for the format and the four acceptance rules. Disclosure: the maintainer of this list also builds some of the tools listed in it (TranscriptAPI, StayingAPI, Zillapi). Our own entries follow the exact same format and bar as everyone else's, appear at most once per category, and we never reject a competing entry to protect ours — see the disclosure section of CONTRIBUTING.md for the full policy.

Maintained by ZeroPointRepo · list content licensed CC BY 4.0 · Built with crhq.ai

About

A curated directory of 31+ verified Agent Plugins — the open standard from OpenAI, Amazon, Microsoft, GitHub, Cursor & Vercel that bundles Agent Skills + MCP servers into one portable package.

agentplugins.codes/

Topics

agent-pluginsagent-skillsai-agentsawesomeawesome-listclaudecursorgithub-copilotllm-toolsmcpopenaivscode

Resources

Readme

Contributing

Contributing
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors