MCP tools are great for giving agents the ability to do things — query your database, call your API, trigger an action. But tools don't tell an agent how to use them well. That's what skills are for.

What are skills?

A skill is a reusable prompt template attached to your MCP server. When an agent connects to your server, it sees your tools — and it also sees your skills. The agent can use a skill to get a pre-written prompt, with any placeholders already filled in, ready to submit.

Skills are exposed via the MCP prompts primitive: prompts/list to enumerate them, and prompts/get to retrieve a rendered prompt with arguments substituted. Any MCP-compatible client — Claude, Codex, Cursor, Gemini — picks them up automatically.

The problem skills solve

Without skills, every agent that connects to your MCP has to figure out how to use your tools from scratch. A tool named search_records with a query field is clear enough, but an agent still has to decide:

What to search for
How to phrase the query
What to do with the results

If you have an established way to use your MCP effectively — and you probably do — there's no way to encode that knowledge in the server itself. Your documentation lives somewhere else, your best prompts are in a Notion doc or someone's head, and every new user starts from zero.

Skills solve this by moving your best practices into the server where they belong.

What skills look like

A skill has a name, an optional title and description, a prompt template, and a list of arguments. The prompt template uses {{argument_name}} placeholders that get filled in when an agent calls the skill.

Here's a simple example:

{
  "name": "summarize_records",
  "title": "Summarize Recent Records",
  "description": "Fetch and summarize the most recent records from your database",
  "prompt": "Use the search_records tool to find records matching '{{topic}}' from the last 7 days. Summarize the key themes in 3 bullet points.",
  "arguments": [
    { "name": "topic", "description": "The topic to search for", "required": true }
  ]
}

When an agent calls prompts/get with { "name": "summarize_records", "arguments": { "topic": "billing" } }, it gets back:

Use the search_records tool to find records matching 'billing' from the last 7 days. Summarize the key themes in 3 bullet points.

A ready-to-use prompt that incorporates your context, your tools, and your preferred workflow.

Why this matters for agent-agnostic MCP servers

The biggest benefit of skills is that they work across every agent that speaks MCP. You don't write a prompt for Claude and a separate one for Codex. You write it once, attach it to your server, and every compatible client picks it up.

This is especially useful when you're building an MCP server for other people to use. You're not just handing them a set of tools — you're handing them a set of tools and the playbook for how to use them. The skill becomes the interface; the tool is the implementation detail.

Skills vs. system prompts

You might be thinking: can't you just write a good system prompt? Yes — but system prompts live in the client configuration, not the server. They're not portable, they require each user to set things up themselves, and they drift over time.

Skills live on the server. When you update a skill, every agent that connects gets the updated version automatically. No syncing required.

Getting started with skills

You can add skills to any Shredly MCP server from the dashboard or via the Shredly MCP itself:

Add a skill named "summarize_records" to my MCP server. The prompt should...

Skills have no effect on tools/call auth or billing — they're just prompt templates. There's no reason not to add them.

The best place to start is your most common workflow. What does a new team member always ask when they first connect to your MCP? Write that down, turn it into a skill, and let the server answer it.