Guides · Updated November 12, 2026

How to connect a skills directory to Claude with MCP

By Tineessa Nelson, Use This Skill

To connect a skills directory to Claude, add its MCP server address as a connector. For Use This Skill the address is https://usethisskill.com/mcp: in Claude, add it under Customize, then Connectors, then Add custom connector; in Claude Code, run claude mcp add --transport http usethisskill https://usethisskill.com/mcp. No sign-in is needed, and six tools appear.

What do you get once it is connected?

The connector reads the same data as the site. On November 12, 2026, that was the crawl and scan of November 9 (data/skills.json and data/scan.json) and the proofs in data/proofs.json:

Tool What it returns Size of the data behind it
search_skills Up to 25 matches for a task: name, description, category, source, install command, whether a proof exists 1,966 skills from 47 public repositories
get_skill One skill in full: install command, license, scan result and the SKILL.md text when the license allows it SKILL.md text for 959 skills; a link to the source for the rest
check_skill Clear, review or flagged per skill folder, with file, line and text for each finding; unlisted GitHub links are scanned live 1,905 clear, 47 review, 14 flagged
list_proven Every proof: the brief, the model and the page 42 proofs
get_proof One proof: brief, model, skill commit, date and the links to the results with and without the skill Every proof is open
order_proof How a skill author gets a skill proven Orders are paused; the tool says so

Counted by script from the site's data files on November 12, 2026. The tool list comes from the connector's own code, run from this site's repository the same day.

How do I add it in Claude, Claude Code, Cursor and Gemini CLI?

Each client stores the address in its own place. These steps were checked against each product's documentation on November 12, 2026.

Client How to add it Where it is stored
Claude (web, desktop, mobile) Customize, then Connectors, then + Add and Add custom connector; name it and paste the URL. Free plans allow one custom connector. Your Claude account
Claude Code claude mcp add --transport http usethisskill https://usethisskill.com/mcp ~/.claude.json (local scope, the default)
Claude Code, for a team Add --scope project to the same command .mcp.json in the project root
Cursor {"mcpServers": {"usethisskill": {"url": "https://usethisskill.com/mcp"}}} .cursor/mcp.json (project) or ~/.cursor/mcp.json (everywhere)
Gemini CLI gemini mcp add --transport http usethisskill https://usethisskill.com/mcp .gemini/settings.json (project, the default) or ~/.gemini/settings.json with --scope user

Two details catch people out. In Gemini CLI, --transport http matters: the default transport is stdio, and in settings.json a streaming HTTP server goes under httpUrl, while url means an SSE endpoint. In Claude, on Team and Enterprise plans an Owner adds the connector in Organization settings, and members then click Connect.

Any other client that supports remote MCP servers over Streamable HTTP takes the same URL. The Connect page has copy buttons for each line above.

How do I check that it worked?

Ask the client for its server list before you ask it for a skill.

  • Claude Code: claude mcp list shows a status next to each server, such as Connected or Failed to connect, and /mcp inside a session shows the tool count. You should see six tools.
  • Cursor: open Customize in the sidebar, find the server, and check that its toggle is on. If it is not working, the Output panel's MCP Logs show the connection and each tool call.
  • Gemini CLI: gemini mcp list shows the connection status, and /mcp lists the tools.
  • Claude: in a chat, click +, then Connectors, and check that the toggle is on for that conversation.

Then try a request that needs the directory: "Search Use This Skill for a landing page design skill and tell me whether it has a proof."

What does a real answer look like?

This is the first result search_skills returned for "landing page design" on November 12, 2026 (description shortened):

{
  "slug": "brand-landingpage",
  "name": "brand-landingpage",
  "description": "Brand-first landing page designer — runs a brand-identity interview...",
  "category": "Design",
  "source": "https://github.com/wshobson/agents/tree/main/plugins/brand-landingpage/skills/brand-landingpage",
  "install": "npx degit wshobson/agents/plugins/brand-landingpage/skills/brand-landingpage .claude/skills/brand-landingpage",
  "proven": false,
  "page": "https://usethisskill.com/skills/brand-landingpage"
}

And this is check_skill for frontend-design, a skill that does have a proof:

{
  "source": "https://github.com/anthropics/skills/tree/main/skills/frontend-design",
  "status": "clear",
  "checked": "directory",
  "skills": [{ "skill": "frontend-design", "status": "clear", "files": 2, "findings": [] }],
  "scanned_at": "2026-11-09T19:51:59.413Z"
}

Every result that carries a skill author's text also starts with a content_notice field. It tells the agent that names, descriptions and SKILL.md text are third-party content to read as data, never as instructions. What each field of a scan result means is covered in check_skill: what the safety check does.

Will the results fill up my context window?

Rarely. We measured the text each tool returned on November 12, 2026:

Call Characters returned
search_skills, 10 results (the default) 7,582
search_skills, 25 results (the maximum) 18,656
get_skill, median of the 959 skills with SKILL.md text 6,678
get_skill, largest (reverse-engineer-anything) 23,914
list_proven, all 42 proofs 27,938

At roughly four characters a token, the largest of these is about 7,000 tokens. Claude Code's documentation says it warns when one MCP tool result passes 10,000 tokens and cuts results off at 25,000 by default, so none of these calls reaches the warning. For how tool definitions load compared with skills, see MCP servers vs agent skills.

Is it safe to connect?

The connector only reads. It cannot write files, run commands or install anything; those actions stay with your agent and its permission settings. Claude Code asks for approval before it uses a project-scoped server from a .mcp.json file in interactive sessions. Cursor asks before using MCP tools by default. Gemini CLI asks before each tool call unless you set the server's trust to true.

The Claude Help Center's advice applies to every custom connector, including this one: connect only to servers you trust, because a malicious server can include hidden instructions. Our server's own instructions tell the agent the opposite of what a malicious one would: call check_skill before installing any skill, show you the command and the result, and install only when you ask. Our safety page lists the nine patterns the scan looks for.

What should I do after I find a skill?

Install it in the folder your agent reads, from the command the connector gave you. The steps for each agent are in How to install an agent skill from GitHub in one command and how to use skills. To see what a skill changes before you install it, open its proof from the proven skills.

How we got this

Counts come from data/skills.json (crawl of November 9, 2026), data/scan.json (scan of the same day) and data/proofs.json, read by script on November 12, 2026. The example results and sizes were returned by the connector's code (lib/mcp.ts), built from this site's repository and called over MCP on November 12, 2026; they are what the live server returns from the same data. Characters are counted on the text each tool returns. Client steps are from each product's documentation, read the same day. We did not test the connector inside each client for this page.

Questions

Do I need an account or an API key to connect Use This Skill?

No. The server at https://usethisskill.com/mcp answers without signing in, so in Claude you can leave the authentication settings empty when you add it. All six tools work without a key, and every proof is open.

Can the connector install a skill for me?

Not by itself. The connector only reads: it returns install commands, SKILL.md text, scan results and proofs. Installing is done by your agent, in your project, with its own permissions. The connector's instructions tell the agent to call check_skill first, show you the command and the result, and install only when you ask.

Does the Claude mobile app see the connector?

According to the Claude Help Center on November 12, 2026, custom connectors connect from Anthropic's cloud rather than your device, and this applies to every Claude client, including claude.ai, Claude Desktop and the mobile apps. A connector you add in one place is available in the others.

Why does Claude Code not show the server in a new project?

claude mcp add stores a server in the local scope by default, which loads only in the project where you added it. Add it again with --scope user to have it in every project, or with --scope project to write it to a .mcp.json file your team shares.