AI Ecosystem Organizer

A local catalog of Skills and MCP servers across the AI clients on your machine (Grok, Hermes, OpenCode, Cursor, Claude, Kimi, Pi, and ~/.agents/skills).

It scans your configs, groups skills into packs (parent + children), and shows how to invoke each tool. Secrets are never stored: only environment variable keys.

Organizer is also agent-facing: it ships its own skill (organizer-catalog) so an AI can configure and operate the catalog, and its own MCP server so the agent can list, inspect, classify, and refresh items without opening the UI.

Why this exists

Skills and MCP servers do not live in one place. Each client has its own folders and config files (~/.agents/skills, ~/.grok, ~/.cursor, Hermes, OpenCode, …). After a few installs you have hundreds of tools and no map.

Packs make that worse: npx skills add owner/repo flattens a whole GitHub repo into sibling folders. A pack like bug-bounty is one parent plus dozens of hunt-* children, but on disk they look unrelated. An agent then has to guess which skill to load from a giant list of descriptions.

Organizer is useful because it:

• Inventories every skill and MCP it can find on the machine, including how to invoke it
• Groups packs (repo, nested folder, or alias) so a parent shows its children
• Gives agents a real API (MCP + skill) instead of hoping they pick the right SKILL.md from memory
• Keeps secrets out of the catalog (env keys only)

If you only use one client and a handful of skills, you may not need it. It pays off when the same machine has several agents and large skill packs.

Quick path

1. Install Go 1.27+ and Bun (Node 22.12+ also works for the frontend).

2. Clone this repo and install frontend deps:

   git clone https://github.com/josegil1909/skill-organizer.git
   cd skill-organizer
   make install

3. Start API + UI:

   make dev

4. Open http://localhost:4321. Keep the backend on http://localhost:8080 running — without it the UI shows a sample catalog, not yours.

What you get

Surface	What it does
Catalog UI	Search, filter by client / category / pack, inspect invocation
Packs	Groups a skills.sh repo, a nested folder (Hermes skills/erp/…), or aliases like bug-bounty → hunt-*
Agent skill	organizer-catalog tells an AI how to wire MCP, triage unclassified tools, and add taxonomy rules
MCP stdio	First-party server: list, inspect, classify, stats, refresh
REST API	GET /api/items, /api/stats, POST /api/refresh

Agent skill

The repo includes .agents/skills/organizer/SKILL.md (name: organizer-catalog). Copy or symlink that folder into the client's skills directory (for example ~/.agents/skills/organizer or .grok/skills/organizer). Once loaded, an agent can:

• Connect Organizer as an MCP server
• Browse the catalog and explain how to invoke a skill or MCP
• Find unclassified items and add taxonomy rules
• Trigger a rescan after you install new tools

You do not need the UI open for that path. The skill is the playbook; the MCP is the API the agent calls.

MCP server

After make build, point any MCP client at the binary (stdio):

{
  "mcpServers": {
    "organizer": {
      "command": "/absolute/path/to/organizer/bin/organizer-backend",
      "args": ["mcp"]
    }
  }
}

From a clone, without installing the binary:

go run ./backend mcp

Tools the server exposes: organizer_list_items, organizer_get_item, organizer_get_unclassified, organizer_add_taxonomy_rule, organizer_get_stats, organizer_refresh. Full argument lists are in the skill file.

Production build

make build

• Binary: bin/organizer-backend (HTTP on :8080, or mcp for stdio)
• Static UI: frontend/dist/ (still talks to http://localhost:8080)

Override the API port with PORT=8081 make dev is not wired in Make; set PORT when running the backend directly: PORT=8081 go run ./backend.

Security

• Bind is local. Do not expose :8080 or :4321 to the internet.
• The scanner records env keys only, never values.
• Taxonomy rules you add persist under ~/.config/organizer/taxonomy.json.

License

MIT. See LICENSE.