Copying a skill into a project works. Letting the agent find the right one itself works better. emdly runs an MCP server at https://emdly.com/mcp; once Claude knows about it, "find me a skill for scoring Jira tickets" becomes a tool call, not a browser tab.
Setup
Claude Desktop and claude.ai: Settings → Connectors → Add custom connector. Name it emdly, paste the URL, and pick Required when the server asks as the authentication. Add. The connector shows up under the tools icon in every chat.
Claude Code:
claude mcp add --transport http emdly https://emdly.com/mcp
Add --scope user if you want it in every project rather than just this one.
That is the whole setup. No API key, nothing to paste.
The public tools
Searching and reading never need an account:
search_skills(query, category?)— ranked results with owner, description and install counts.get_skill(owner/skill, version?)— the Markdown body, latest or pinned.install_skill(owner/skill)— the body plus the path to save it under (./skills/name.md), so Claude can write the file itself.
Ask Claude for "a skill that turns merged PRs into a changelog" and it will search, read the result and offer to install it. Every load counts as an install for the author.
Your Brain, after a sign-in
The Brain tools — brain_list, brain_search, brain_get, brain_write — are private. The first time Claude calls one, emdly answers "sign in" and Claude opens your browser: sign in with Google or GitHub, approve the app, done. From then on Claude can read what you have written down and add to it: "save that we ship on Tuesdays to my engineering notes" becomes a versioned edit in your Brain.
Two things worth knowing. Every write keeps the previous version, so nothing Claude does is destructive. And the Brain refuses anything that looks like a credential — a Slack token pasted into a notes file is bounced, not stored.
Taking it away again
Settings → Connected apps lists every client you approved, with a Disconnect button that revokes its access on the spot. Clients that cannot open a browser (CI, a server-side agent) can use a personal token from Settings instead, sent as a bearer header.
The full tool reference and the token setup are in the docs.