v0.5.0 · Apache-2.0 · Free & open source

One GitHub repo.
Every AI agent.

Stop copy-pasting SKILL.md files into ~/.claude, ~/.cursor, ~/.codex. Skills live in one repo you own. Agents fetch them on demand over MCP — no startup-token tax, no drift.

Install in one command View on GitHub

uvx skills-registry init · needs gh + uv

~ / skills-registry — zsh

$ uvx skills-registry init# Verifying prerequisites…✓ gh authenticated as anand-92✓ Downloaded skills-registry (darwin/arm64) — 4.2 MB✓ Handoff → skill-registry bootstrap# Scanning ~/.* for existing skills…found 11 skills in ~/.claude/skillsfound 6 skills in ~/.cursor/skillsfound 3 skills in ~/.factory/skills# Creating registry repo…repo: anand-92/my-skills (private)✓ Created repo · pushed 20 skills✓ Wired 7 agents · wrote SKILL.md pointersDone. Paste this into your MCP client:# Claude Code / Claude Desktop / Cursor / VS Code — mcp.json{ "mcpServers": { "skill-registry": { "command": "skill-registry-mcp" } }}# Codex — ~/.codex/config.toml[mcp_servers.skill-registry]command = "skill-registry-mcp"$

50+

AI tool dot-folders auto-detected at bootstrap

MCP tools exposed — list_skills, get_skill, publish_skill

SSH keys, git configs, or shell PATHs required

Command to install & wire up every agent

The problem

Every AI tool wants its own skills folder. Edit once, sync N times.

Today every AI coding tool keeps a local skills folder. Same skill, copy-pasted into N dot-folders. Worse — every skill in every folder gets auto-loaded into the agent's startup context, whether the current task needs it or not. You pay tokens for skills you'll never use this conversation.

Before · local dot-folders

Duplication, drift, token bloat

~/.claude/skills/code-review/SKILL.md
~/.cursor/skills/code-review/SKILL.md
~/.codex/skills/code-review/SKILL.md
~/.factory/skills/code-review/SKILL.md
…and 46 other agents you forgot about
All loaded on every startup. Every conversation.

After · skills-registry

One repo. Fetched on demand.

anand-92/my-skills/code-review/SKILL.md
Every agent points to the same repo
Edit once · branch · PR · fork · restore
Pointer file in each agent's dot-folder (~200 bytes)
Real skill fetched only when needed
Cache invalidated on tree SHA change

How it works

Single source of truth. Six deliberate constraints.

Every design decision falls out of one observation: desktop MCP clients spawn the server with a stripped environment — no shell PATH, no SSH agent, no git config user.email. So we don't depend on any of them.

Fetched on demand

Tiny pointer file in each agent's dot-folder. The actual skill is downloaded the moment get_skill(slug) is called — and not before.

50+ tools recognized

Claude Code, Claude Desktop, Cursor, Codex, Windsurf, Goose, Factory, VS Code/Copilot — all detected at bootstrap, universal ones pre-selected.

gh-only GitHub I/O

No git shell-out. No SSH. No embedded HTTP. Every call routes through the user's authenticated GitHub CLI via the Git Data API.

Tree-SHA cache

Skills cached in ~/.cache/skills-mcp/skills/. Cache key is GitHub's tree SHA — force-pushes and subtree changes invalidate correctly.

Path-traversal hardened

publish_skill rejects .. segments and backslash traversals. Per-file size cap (2 MiB). Identical validation in Python & Go.

Git, but for skills

The registry is a real GitHub repo. Branch it, PR it, fork a teammate's, restore old versions. Apache-2.0 — yours forever.

Comparison

Why not just keep using local folders or a dotfiles repo?

Capability	Local dot-folders	Dotfiles repo	skills-registry
One home for every agent	duplicated	yes	yes
Fetched on demand (no startup tokens)	no	no	yes
Versioned + branchable	no	yes	yes
Works in every MCP client	partial	no	yes
Share / fork between users	no	clunky	clone the repo
No shell or SSH config needed	yes	no	yes

Architecture

Three deliverables. One source repo.

Python bootstrap + Python MCP server + Go TUI manager. The Python wheel ships both Python entry points; the Go binary is downloaded from GitHub Releases on first run.

skills-registryPython 3.10+

Thin bootstrap. Verifies gh, downloads the Go CLI, execs it. One command: skills-registry init.

PyPI wheel · entry point

skill-registry-mcpPython 3.10+

FastMCP server exposing list_skills, get_skill, publish_skill over MCP stdio. Same wheel, second entry point.

PyPI wheel · MCP stdio · FastMCP 3.x

skill-registryGo 1.24+

Charmbracelet TUI manager. Commands: bootstrap, list, get, sync, add, publish.

GitHub Releases · darwin/linux/windows × amd64/arm64

MCP surface

Three tools. Any MCP-aware agent can call them.

list_skills

Enumerates every skill in the registry. Returns a markdown table with slug, name, description, and the URI to fetch.

read

get_skill(slug)

Downloads one skill into the local cache. Returns the absolute path. Cache invalidated on tree-SHA change.

read

publish_skill(name, …)

Publishes a skill via the GitHub Git Data API. Accepts files mapping or local_folder. Returns commit SHA. 3 retries on conflict.

write

Users don't call these directly. They just say "what skills do I have?" or "use the code-review skill on this PR" — the agent picks the right tool.

// Claude Code / Claude Desktop / Cursor / VS Code — mcp.json{"mcpServers": {"skill-registry": {"command": "skill-registry-mcp"}}}

# ~/.codex/config.toml[mcp_servers.skill-registry]command = "skill-registry-mcp"

# What the agent ends up doing under the hooduser: "use my code-review skill on this PR"agent: get_skill("code-review")→ /Users/you/.cache/skills-mcp/skills/code-review/→ reads SKILL.md→ follows the skill's instructions

skills-registry init prints the platform-correct snippet for you.

Install

One command. Five things happen.

Verify gh is authed
Exits with code 3 if missing, code 4 if not logged in. Run gh auth login first.
Download the Go CLI
Pulls the matching tarball from GitHub Releases into ~/.local/bin (or $SKILLS_BIN_DIR).
Hand off → skill-registry bootstrap
Scans ~/.* for known dot-folders. Pre-selects universal ones (factory, codex). You confirm the multi-select.
Create the registry repo
Calls gh repo create. Pushes every found skill via the Git Data API. Writes SKILL.md pointer files into each agent.
Print the MCP snippet
Platform-correct JSON / TOML you paste into your MCP client. Restart the client. Done. Re-running init is safe — idempotent.

# No system Python needed — uvx handles it$ uvx skills-registry init# Then in your MCP client:$ cat ~/.config/claude/mcp.json

# Install persistently — MCP clients won't depend on uvx cache$ uv tool install skills-registry# Installs both entry points:#   skills-registry            (bootstrap CLI)#   skill-registry-mcp         (MCP stdio server)$ skills-registry init

# Classic pip — needs Python 3.10+ already on PATH$ pip install skills-registry$ skills-registry init# Prerequisites checked at runtime:#   - gh        (https://cli.github.com/)#   - gh auth login already run

Prereqs: gh + uv. macOS, Linux, Windows (best-effort).

CLI reference

The skill-registry binary

Charmbracelet TUI for day-to-day management. Same Git-Data-API publish flow as the MCP server, mirrored in Go.

Command	What it does
skill-registry bootstrap	First-run setup. Idempotent — safe to re-run.
skill-registry list	Fuzzy-filterable TUI of every skill in your registry. Press / to search, Enter to preview.
skill-registry get <slug>	Download one skill into ./skill-registry/<slug>/.
skill-registry sync	Push local skills sitting in .claude/skills, .cursor/skills, etc. that aren't yet in the registry.
skill-registry add <owner/repo>	Clone a teammate's registry. Multi-select which of their skills to pull into your own.
skill-registry publish <path>	Publish a single local skill folder. Path-traversal validated. 2 MiB per-file cap.
skill-registry --version	Print version. Current: 0.5.0.

Override the registry per-process with SKILLS_REGISTRY=owner/repo — useful for browsing a teammate's read-only.

Free, open, and pre-1.0. Try it today.

Apache-2.0. No accounts. No telemetry. No paid tier — ever. Built by anand-92 as an open-source dev tool. Star the repo or file an issue if it breaks.

uvx skills-registry init ★ Star on GitHubMCP surface stable · internals may shift between minor versions

One GitHub repo.
Every AI agent.

Every AI tool wants its own skills folder. Edit once, sync N times.

Duplication, drift, token bloat

One repo. Fetched on demand.

Single source of truth. Six deliberate constraints.

Fetched on demand

50+ tools recognized

gh-only GitHub I/O

Tree-SHA cache

Path-traversal hardened

Git, but for skills

Why not just keep using local folders or a dotfiles repo?

Three deliverables. One source repo.

Three tools. Any MCP-aware agent can call them.

One command. Five things happen.

Verify `gh` is authed

Download the Go CLI

Hand off → `skill-registry bootstrap`

Create the registry repo

Print the MCP snippet

The skill-registry binary

Free, open, and pre-1.0. Try it today.

One GitHub repo.Every AI agent.

Every AI tool wants its own skills folder. Edit once, sync N times.

Duplication, drift, token bloat

One repo. Fetched on demand.

Single source of truth. Six deliberate constraints.

Fetched on demand

50+ tools recognized

gh-only GitHub I/O

Tree-SHA cache

Path-traversal hardened

Git, but for skills

Why not just keep using local folders or a dotfiles repo?

Three deliverables. One source repo.

Three tools. Any MCP-aware agent can call them.

One command. Five things happen.

Verify gh is authed

Download the Go CLI

Hand off → skill-registry bootstrap

Create the registry repo

Print the MCP snippet

The skill-registry binary

Free, open, and pre-1.0. Try it today.

One GitHub repo.
Every AI agent.

Verify `gh` is authed

Hand off → `skill-registry bootstrap`