skill-finder

One skill to unlock 89,000.

An agent skill that gives Claude Code, Codex CLI, and any skill-compatible agent instant access to a knowledge base of 89,267 expert guides — covering tools, frameworks, cloud platforms, DevOps, ML, databases, and more.

No local setup. No MCP. Just drop in the skill and go.

How It Works

You: "How do I deploy to Kubernetes?"
                ↓
       skill-finder triggers
                ↓
   Agent searches 89K guides via API
                ↓
    Fetches best match (full instructions)
                ↓
   Agent follows the guide to complete your task

The agent experience is seamless — it searches, fetches, and applies the right guide as if it always knew how.

Install

Claude Code

cp -r skill-finder/ ~/.claude/skills/skill-finder/

Codex CLI / Goose / Amp

cp -r skill-finder/ ~/.agents/skills/skill-finder/

That's it. No API keys, no dependencies, no configuration.

What's Inside

skill-finder/
├── SKILL.md              # The skill — agents load this
├── scripts/
│   └── serve.py          # Self-hosted API server (optional)
└── Dockerfile            # Deploy your own instance (optional)

• SKILL.md is all you need. It calls a hosted API at zcheng256-skill-retrieval.hf.space.
• scripts/serve.py is the API server source — self-contained, no MCP dependency. Deploy your own if you want full control.

Knowledge Base

89,267 skills sourced from:

Source	Skills	Examples
LangSkills	72K+	Tool guides, framework workflows, deployment patterns
Anthropic Official	50+	PDF, PPTX, DOCX, algorithmic art, MCP builder
Community	200+	Superpowers, custom workflows
SkillNet	16K+	Research-derived task skills

Covers: Kubernetes, Terraform, Docker, AWS, GCP, React, Python, Rust, PostgreSQL, Redis, Grafana, GitHub Actions, and thousands more.

skill-finder vs skill-retrieval-mcp

This project is the zero-setup companion to skill-retrieval-mcp. Same knowledge base, different integration approach.

	skill-finder	skill-retrieval-mcp
What it is	A single skill file (SKILL.md)	An MCP server
Setup	Copy one file	pip install + download 1.1GB data + register
Runs where	Cloud API (hosted)	Local (your machine)
Requires MCP	No	Yes
Agent support	Any agent with skills + WebFetch	MCP-compatible agents only
Latency	~200ms (network)	<5ms (local)
Custom skills	No (uses hosted DB)	Yes (import your own)
Offline	No	Yes

Choose skill-finder if you want instant access with zero setup. Choose skill-retrieval-mcp if you need local speed, offline access, or custom skills.

Self-Hosting

Want to run your own API? The server is a single Python file with zero MCP dependency:

pip install starlette uvicorn faiss-cpu numpy sentence-transformers huggingface-hub

# Data downloads automatically on first run
python scripts/serve.py

Or deploy via Docker:

docker build -t skill-finder .
docker run -p 7860:7860 skill-finder

Update the API URL in SKILL.md to point to your instance.

API

Endpoint	What it does
GET /api/search?q=...&k=5	Semantic search (returns summaries with scores)
GET /api/skills/{id}	Full skill instructions by ID
GET /api/search/keyword?q=...&limit=10	Keyword search (FTS5)
GET /api/categories	List all categories and counts
GET /api/health	Server status

License

MIT