auto-skill-connector
# auto-skill-connector
An MCP connector that gives Claude Code / Claude Desktop access to a database
of ~200k scraped Claude skills, MCP servers, and plugins. Instead of building
a capability from scratch, Claude can search this database mid-conversation,
read a matching skill's instructions, and follow them immediately — or
install it permanently as a real `/skill`.
It ships two tools:
- **`recommend_skill(task)`** — hybrid full-text + semantic search over the
skill database. Returns the best match's full `SKILL.md` content (or a
short list to pick from, if a few skills fit equally well).
- **`install_skill(url, name?)`** — downloads a skill's `SKILL.md` and saves
it to `~/.claude/skills/<name>/SKILL.md`, so Claude Code can invoke it as a
normal skill from then on, in any project.
Query embedding happens server-side (Supabase Edge Function), so this
connector only depends on `mcp` + `httpx` — no local ML runtime to install.
## Install
### Claude Code
```
claude mcp add auto-skill --scope user -- uvx --from git+https://github.com/neelavalareddy/auto-skill-connector auto-skill-mcp
```
### Claude Desktop
Add this to your `claude_desktop_config.json` (Settings → Developer, or find
it directly — on Windows it's usually under
`%APPDATA%\Claude\claude_desktop_config.json`):
```json
{
"mcpServers": {
"auto-skill": {
"command": "uvx",
"args": ["--from", "git+https://github.com/neelavalareddy/auto-skill-connector", "auto-skill-mcp"]
}
}
}
```
Restart Claude Desktop after editing the config.
Both require [`uv`](https://docs.astral.sh/uv/) installed (`uvx` ships with
it) — no cloning or manual `pip install` needed.
## Automatic skill suggestions (optional, Claude Code)
Want every chat message checked against the database automatically? Add the
included `hooks/skill_suggest.py` as a `UserPromptSubmit` hook: it runs on
each prompt you send, and when a skill matches, Claude is told to fetch and
apply it via `recommend_skill`. It fails open — errors and timeouts never
block or slow your chat.
1. Download [`hooks/skill_suggest.py`](hooks/skill_suggest.py) somewhere
permanent (e.g. `~/.claude/hooks/skill_suggest.py`).
2. Merge this into `~/.claude/settings.json` (use an absolute path on
Windows, e.g. `C:\\Users\\you\\.claude\\hooks\\skill_suggest.py`):
```json
{
"hooks": {
"UserPromptSubmit": [
{
"hooks": [
{
"type": "command",
"command": "python",
"args": ["~/.claude/hooks/skill_suggest.py"],
"timeout": 10
}
]
}
]
}
}
```
## Running it directly
```
git clone https://github.com/neelavalareddy/auto-skill-connector
cd auto-skill-connector
pip install -e .
python mcp_server.py
```
## How it works
The skill database lives in Supabase (Postgres + pgvector), populated by a
separate scraper that continuously crawls GitHub, npm, and the MCP registry
for Claude skills. This repo only contains the read-only connector — search
queries hit a read-only anon key (RLS grants `SELECT` only; no writes are
possible with it).
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: one searches and returns skill content, the other installs a skill from a URL. No overlap in functionality.
Both tools follow a consistent verb_noun pattern: 'install_skill' and 'recommend_skill', making them predictable and easy to distinguish.
With only two tools, the server feels minimal for a skill management system. While it covers the core workflow of finding and installing skills, it lacks tools for listing, updating, or removing skills, which limits its utility.
The server covers the basic find-and-install flow but is missing lifecycle operations like listing installed skills, updating, or removing them. Users may need additional tools to manage their skill set effectively.