Teach Claude new skills 8 skills
Claude apps (web/desktop/mobile): Skills (upload a SKILL.md zip), Connectors (MCP-based, from the Claude directory), and Projects with custom instructions.
Where: claude.ai → Customize → Skills / Connectors; directory at claude.ai/directory
Skills Plugins MCP Tools
🛡 Stay safe Read first
Official first — from the AI maker or the service’s own GitHub org / domain. Read SKILL.md and any scripts before installing. Skills can run code with your permissions. Never paste base64 or “curl … | sh” download-and-run lines from a skill.
Scan installed skills (free scanner by Snyk):
uvx snyk-agent-scan@latest <skills-folder> Copy
Real incidents: ClawHavoc (341+ malicious skills) · Snyk ToxicSkills (76 / 3,984) · fake postmark-mcp
Anthropic built-in document skills 🟢 Official
Create/edit Word, Excel, PowerPoint and PDF files inside Claude.
Skills ★ 197k License: see repo updated 2 days ago
Settings → Capabilities → turn on Code execution and file creation Customize → Skills → enable the Anthropic skills (docx, xlsx, pptx, pdf)
⚠️ Official. Still review what it can access before enabling.
Can run code Reads your files
skill-creator 🟢 Official
Teaches Claude to interview you and write a new skill for your own repeatable workflow.
Skills ★ 197k License: see repo updated 2 days ago
Customize → Skills → enable skill-creator Ask: 'Help me create a skill for <task>'
⚠️ Official. Still review what it can access before enabling.
Can run code
Knowledge-work plugins (sales, finance, legal, support…) 🟢 Official
Role-based bundles of skills + connectors for business teams.
Plugins ★ 36.4k Apache-2.0 updated 1 day ago
In Claude Code/Cowork: claude plugin marketplace add anthropics/knowledge-work-plugins claude plugin install sales@knowledge-work-plugins
⚠️ Official. Still review what it can access before enabling.
Can run code
Custom connector (remote MCP URL) 🟢 Official
Connect any remote MCP server (e.g. https://mcp.exa.ai/mcp, https://mcp.notion.com/mcp).
MCP ★ 7.5k License: see repo updated 5 days ago
Customize → Connectors → Add custom connector Paste the server URL → authenticate
⚠️ MCP servers run with your permissions and can read data you give them; only connect servers you trust.
Can run code Acts on your account
Official MCP Registry 🟢 Official
Directory of published MCP servers (the open catalog run by the MCP project). Browse to find a server for any app.
Tools ★ 7.5k License: see repo updated 5 days ago
Open registry.modelcontextprotocol.io (or API: GET /v0.1/servers?search=<name>) Prefer servers whose namespace matches the vendor's own domain/GitHub org Copy the package/remote URL into your client
⚠️ The registry hosts metadata, not audits — a listing is not a security review.
Reads your files Acts on your account
Upload any custom skill (zip) 🟢 Official
Add a community or self-made skill (folder with SKILL.md).
Skills
Zip the skill folder (folder name = skill name, SKILL.md inside) Customize → Skills → + Create skill → Upload a skill
⚠️ Only upload skills whose files you've read; scripts run in Claude's code sandbox with your files.
Can run code Reads your files
Claude Connectors directory 🟢 Official
One-click connections to Gmail, Google Drive, Notion, Slack, GitHub, Canva, etc.
MCP
Customize → Connectors → Browse Pick the vendor's own connector → Connect → sign in
⚠️ Connectors act with your account permissions; prefer read-only where offered.
Acts on your account
awesome-claude-skills (ComposioHQ list)
Big curated list of community skills for writing, data, media, dev.
Tools ★ 79.7k License: see repo updated 2 months ago
Pick a skill → open its repo → download folder → zip → upload as custom skill
⚠️ Skills can include scripts. Snyk's ToxicSkills study (Feb 2026) scanned 3,984 skills from ClawHub and skills.sh and confirmed 76 malicious ones; scan with `uvx snyk-agent-scan@latest <skills-folder>` before trusting community skills.
Can run code Reads your files
No skills match these filters. Show all
Stars, license and last update come from GitHub (17 Nov 2026). Commands are copied from official docs and are not tested on a real machine yet. Access tags are our summary — always check the source. Skills for every AI →