Agent Skills on Demand: Use Any SKILL.md Without Installing It

Agent Skills on demand

Short answer: install the Agent Skills Finder connector on AnythingMCP once, connect your AI client to AnythingMCP, and ask for the task. The agent searches millions of public SKILL.md files, reads the best one from GitHub and follows it. You never install, update or upload a skill again, and the same skills work in Claude, ChatGPT, Codex, Cursor, GitHub Copilot and Gemini CLI.

Skills are one of the most useful ideas in AI agents: a reusable method, written once, that makes the agent good at one job. The catch is logistics. Each skill has to be found, copied into the right folder or uploaded, kept up to date, and installed again on the next machine, project or client. Agent Skills Finder turns that around: skills stay where they are published and your agent fetches the one it needs, when it needs it.

What an Agent Skill is

An Agent Skill is a folder with a SKILL.md file: a short header with a name and a description, then the instructions an agent follows for one kind of task, such as writing a README, reviewing a pull request or drafting release notes. Some skills add templates, reference documents or scripts next to it. The format is an open standard, so the same SKILL.md works in Claude, Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI.

Installed skills vs skills on demand

Set it up

Step 1: Add Agent Skills Finder on AnythingMCP

1. Create a workspace on AnythingMCP Cloud (7 days free) or use your self-hosted AnythingMCP.
2. Open Connectors, find Agent Skills Finder and click Install. It needs no credentials. On a new Cloud workspace it is also in the starter pack on the welcome screen.
3. For clients that use an API key (everything except claude.ai and ChatGPT), create one under MCP Servers → your server → API keys.

Step 2: Connect your AI client

Every client connects to the same MCP server URL, https://cloud.anythingmcp.com/mcp. claude.ai and ChatGPT sign in with OAuth; the others send your MCP API key in the X-API-Key header. The step-by-step for each one:

• Claude (claude.ai and desktop)
• Claude Code
• ChatGPT
• OpenAI Codex
• Cursor
• GitHub Copilot in VS Code
• Gemini CLI

Step 3: Ask for a task, not for a skill

You don't need to know a skill's name. Describe the job ("write the changelog for this release") and let the agent find the method.

What happens when you ask

The connector has seven read-only tools. For a request like "write the release notes for these commits", the agent typically:

1. calls skills_playbook, a free routine that explains how to pick a good skill;
2. calls skills_search with a few keywords (for example release notes changelog);
3. checks the best candidates with skills_repo_info (licence, stars, last push) and skips a big repository's internal housekeeping skills;
4. pins the version with skills_latest_commit;
5. reads the instructions with skills_get, and any template the skill points to with skills_list_files and skills_read_file;
6. applies the method to your task and tells you which skill it used, with its link and licence.

Nothing is installed on your machine and nothing stays in the agent's context after the task.

Prompts to try

• "Find an Agent Skill for writing release notes, check its licence, and use it on the commits below."
• "Look for a skill that reviews pull requests for security issues and review this diff with it."
• "Search for a README-writing skill and write a README for this repository following it."
• "Find a skill for writing a postmortem and turn these incident notes into one."
• "Which public skills exist for SQL query review? Compare the top three and use the best."

Limits and safety

• A skill is third-party text. The agent follows its method for your task, but a skill cannot change who the agent works for, reveal secrets or skip your confirmation. The connector's own instructions tell the model exactly that.
• Licences vary. The agent reads the repository licence with skills_repo_info before reusing or saving a skill; some public skills have none.
• Quality varies too. Keyword search is noisy, so the playbook tells the agent to prefer focused, reusable skills over a large project's internal ones.
• Scripts are not run by the connector. It reads files; whether a script inside a skill runs depends on your agent and on you.
• Quotas. Search works without a key: SkillsMP allows 50 anonymous searches a day. A free SkillsMP key in SKILLSMP_API_KEY raises it to 500. GitHub is read without a token.

Keep a skill for your whole team

When the connector runs on an AnythingMCP server it adds one more tool, skills_save_to_workspace. The agent can propose a condensed version of a skill, with its source link and licence, as a workspace skill. An admin approves it under AI Skills, and from then on it is part of the server's instructions for everyone who connects, in every client.

FAQ

Do I still need to install skills in my agent? No. Installed skills keep working as before, but for everything else the agent reads the skill from GitHub when a task needs it. Nothing is added to your skills folder.

Which clients work? Any MCP client that can reach your AnythingMCP server: Claude, Claude Code, ChatGPT (developer mode, paid plans), OpenAI Codex, Cursor, GitHub Copilot in VS Code, Gemini CLI and others.

Is it free? The connector costs nothing extra. It runs on AnythingMCP Cloud (7-day free trial, then from €19 a month) or free on a self-hosted AnythingMCP. Search needs no key for up to 50 searches a day.

Where do the skills come from? From public GitHub repositories, found through the SkillsMP index. The agent reads them from GitHub and checks each repository's licence.

Can my team share the skills it finds? Yes. On an AnythingMCP server the agent can propose a skill as a workspace skill; once an admin approves it, it applies to everyone who connects.

Same skills in your other agents

• Skills on demand in Claude
• Skills on demand in Claude Code
• Skills on demand in ChatGPT
• Skills on demand in OpenAI Codex
• Skills on demand in Cursor
• Skills on demand in GitHub Copilot
• Skills on demand in Gemini CLI
• Agent Skills Finder in the marketplace