Connect Skillsstore to Claude.

Skillsstore is a workspace library of curated skills — Markdown documents capturing the playbooks, brand-voice rules, code-review checklists, and house standards a team has codified. This guide walks through installing Skillsstore as a Custom Connector inside claude.ai (web) and Claude Desktop, so Claude can browse and read those skills directly during a conversation.

What you get

• Three read-only tools — list_skills, search_skills, load_skill — exposed over the standard Model Context Protocol.
• Per-user OAuth: each member of your workspace authorises the connector with their own session. No shared API keys, no credentials in chat.
• Folder-scoped access: when you authorise the connector you choose which folders Claude can see. The rest of the workspace stays private.
• Read-only by design: the connector cannot create, edit, or delete skills — that all happens in the dashboard.

1 — Add the connector in Claude

In claude.ai, go to Settings → Connectors → Add custom connector. Paste this URL and confirm:

https://skillsstore.ai/api/mcp

In Claude Desktop, open Settings → Connectors → Add custom connector and paste the same URL. The flow is identical to claude.ai — no JSON file to edit.

2 — Authorise

Claude opens a browser window pointing at https://skillsstore.ai/cli. Sign in to your Skillsstore workspace, then choose what to expose:

• Entire workspace — every published skill in your workspace is visible to Claude. New skills are picked up automatically (no reconnection needed).
• Specific folders — only the folder subtrees you tick are exposed. Skills outside those folders stay hidden.

Click Authorize. The window closes and Claude confirms the connector is active.

3 — Use it

Ask Claude what skills you have, or ask for help on a topic your workspace has codified. Claude will surface the relevant skills via list_skills / search_skills and load any one of them with load_skill.

Example openers that work well:

• “What skills do I have in Skillsstore?”
• “Use my code-review skill on this PR.”
• “Apply our brand-voice skill to this email draft.”

Tools exposed

All three are flagged readOnlyHint: true; the connector never mutates workspace state.

Authentication

Skillsstore implements OAuth 2.1 with PKCE and Dynamic Client Registration (RFC 7591), the modern flow Claude Connectors uses. Discovery endpoints follow the MCP authorization spec:

• /.well-known/oauth-protected-resource (RFC 9728)
• /.well-known/oauth-authorization-server (RFC 8414)
• /api/oauth/register for Dynamic Client Registration
• /api/oauth/token for the authorization-code grant

Access tokens are workspace-scoped and tied to the authorising user. Revoke them at any time from /dashboard/connect.

Privacy & security

• Skill content is hosted in the EU (Vercel Blob, Frankfurt region) and stored content-addressed by SHA-256.
• Tokens are stored as SHA-256 hashes in our database — the raw token only exists in Claude's connector store.
• We never train AI/ML models on workspace content.
• Full details: Privacy Policy · Terms of Service.

Troubleshooting

• Claude shows generic answers instead of using my skills. Mention "my Skillsstore skills" explicitly in the first message of a new chat — Claude caches tool descriptions per session and a clean prompt usually nudges it to call list_skills first.
• I added skills but Claude can't see them. New skills are picked up automatically, but Claude's tool list is cached per session — start a new conversation. If a folder is scoped, make sure the new skill lives inside one of the authorised folders.
• Authorization failed. Disconnect the connector in claude.ai, reconnect, and re-authorise from /dashboard/connect.

Support

Questions or issues: contact@skillsstore.ai.