New to AI plugins? Start with the job, not the list
Pick what you want your agent to do — review code, debug and test, connect tools like GitHub or a database, browse the web, write docs, or stay safe — and HOL Registry shows the highest-trust plugins for that job. Each listing shows what is inside (skills, MCP servers, apps), a trust score from identity, activity, verification, and security signals, and a copy-ready install command. A scan is not a safety guarantee.
Maker: HOL · Updated 2026-10-21
Your first install, in three steps
- Pick a job below, or open the full catalog, and choose your agent at the top: Claude Code, Codex, Cursor, OpenCode, Antigravity CLI, Grok, Kimi Code, or an MCP client.
- Copy the install command from a card, or open the plugin for its full guide, security signals, and what is inside.
- Run the command, restart your agent session, and confirm the skill, MCP server, or app shows up before you rely on it.
Pick the job
Top picks by HOL trust score. Scores are signals, not a safety guarantee.
Ship & review code
Pull-request reviews, CI checks, release chores, and code-quality passes your agent runs for you.
Debug & test
Reproduce failures, read logs, run the test suite, and explain what broke.
Connect my tools
Give your agent live access to GitHub, Slack, a database, or an internal API.
Browse & research
Search the web, open pages, and summarize what matters into your session.
Write docs & content
READMEs, changelogs, release notes, and posts drafted from the work itself.
Stay safe
Security scans, secret checks, and policy guardrails around what your agent runs.
Getting started questions
What is an AI plugin, in plain terms?
A plugin is a package that extends an AI coding agent. In HOL Registry a plugin can contain skills (instructions and scripts the agent loads on demand), MCP servers (live tool connections the agent can call), and apps (bundled interfaces or services). The catalog card on each listing shows which of those it contains.
What should my first plugin be?
Start with the job you repeat most: reviewing pull requests, debugging failures, or connecting a tool you use daily. Open the matching shortlist on this page, pick a listing with a strong trust score and a publisher you recognize, and copy its install command. You can remove a plugin later; starting small beats installing ten at once.
How do I install a plugin?
Pick your agent in the catalog header (Claude Code, Codex, Cursor, OpenCode, Antigravity CLI, Grok, Kimi Code, or an MCP client) and every card shows the install command for that agent. Copy it from the card, or open the plugin page for the full per-runtime guide. Commands are generated from the plugin’s declared ecosystems and repository.
How do I know a plugin is safe to install?
Compare the HOL trust score, the security score, and owner verification on the listing, then read the scanner-backed signals on the plugin page. A scan is not a safety guarantee, and HOL does not certify every plugin forever. Treat scores as comparison evidence and review the source before installing third-party code.
What is the difference between a skill, an MCP server, and an app?
A skill teaches your agent a workflow (instructions plus optional scripts). An MCP server connects your agent to a live tool or data source it can call. An app is a bundled interface or service that runs alongside the agent. One plugin can ship all three; the catalog filters let you browse by kind.
Limitations
- Shortlists are ranked by HOL trust score and catalog signals, not hands-on testing.
- A scan is not a safety guarantee. Review a plugin’s security signals before installing.
- Install commands differ by agent; the plugin page carries the full per-runtime guide.
- HOL does not replace each vendor’s official marketplace or documentation.
After you install, HOL Guard can apply optional local-first runtime policy for coding agents. It is not required to browse or install.