Agent Plugins specification · v1
The registry for Agent Plugins
An index of open-source skills and MCP servers, packaged as self-contained plugins per the open spec and indexed straight from GitHub.
Registry stats
15
Plugins indexed
24
Skills
12
MCP servers
33
Categories
Featured plugins
Browse all →Recently indexed
- apiscribeOpenAPI-first API documentation. Skills keep specs, prose, and SDK examples in sync; an MCP server serves the spec registry.indexed 50 minutes ago
- deploy-captainRollout planning and incident notes as skills, plus an SSE feed of fleet deploy events so agents can see what shipped where.indexed 50 minutes ago
- frameshiftData-wrangling skills backed by an embedded DuckDB MCP server. Profile a dataset first, then clean it column by column.indexed 50 minutes ago
- argus-scanDependency and secret scanning for agent workflows. Skills audit lockfiles and diffs; the MCP server serves a live advisory feed.indexed 50 minutes ago
- test-pilotGenerates unit tests from real call sites, hunts flaky tests by rerunning suspects in isolation, and exposes coverage data over MCP.indexed 50 minutes ago
Anatomy of a plugin
An Agent Plugin is a self-contained directory: a plugin.json manifest at the root, with skills and MCP servers discovered from fixed locations inside it. Failure boundaries are non-fatal by design — a component that fails to load is skipped, and the rest of the plugin keeps working.
my-plugin/
├── plugin.json # manifest — identity + metadata
├── skills/
│ ├── code-review/
│ │ └── SKILL.md # one directory per skill
│ └── changelog/
│ └── SKILL.md
├── mcp.json # MCP server definitions
└── README.md
Built for agents
Every surface reads from the same index — humans get this site, agents get structured endpoints.
POST /api/mcp
An MCP endpoint agents can call natively to query the registry from inside a session.
MCP endpoint →