# Skills and MCP servers

Source: https://layerwise.app/docs/build-your-app/skills-and-mcp

Teach the agent your conventions with skills, and connect tools from other services through MCP servers.



A **skill** is a set of instructions the agent can follow for a kind of task, such as your stack's conventions or a design process. An **MCP server** gives the agent tools from another service, such as reading a design from Figma or querying an issue tracker.

## Personal and project settings [#personal-and-project-settings]

Skills and MCP servers can be added in two places:

* **Personal** — Open **Account** from the user menu, then choose **Skills** or **MCP Servers**. Personal skills and servers are available in all your projects on this device.
* **Project** — In the project sidebar, open **Settings**, then choose **Skills** or **MCP Servers**. Project skills and server definitions are saved in the project's repository, so everyone who works on the project gets them. Whether each one is turned on is your own setting.

Personal skills and servers also appear in each project's settings. You can turn them on or off there, but edit them from your personal settings.

Changes apply to the next message you send; a turn already running keeps its settings. If the project settings ask you to prepare the workspace, open a conversation in the project first.

## Skills [#skills]

### Install a skill from skills.sh [#install-a-skill-from-skillssh]

1. In **Discover skills**, search for a skill.
2. Choose a result to read its files, or choose **Install** to add it.

### Add your own skill [#add-your-own-skill]

Choose **Add skill**, then one of:

* **Create skill** — Write the skill in Layerwise.
* **Import from GitHub** — Enter a repository or folder URL. Optionally, enter the skill's directory and a branch, tag, or commit. Layerwise copies the skill folder and its files without running any scripts.
* **Import folder** — Choose a skill folder on your computer.

A skill is a folder with a `SKILL.md` file. The file starts with a name and a description:

```md
---
name: release-notes
description: Write release notes in the team's format. Use when asked for release notes or a changelog entry.
---

1. List the user-facing changes since the last release.
2. Group them under Added, Changed, and Fixed.
```

The name uses lowercase letters, numbers, and hyphens, up to 64 characters. The description can be up to 1,024 characters; write it so the agent can tell when the skill applies. An imported skill can be up to 20 MB and 1,000 files, and cannot contain symbolic links.

Project skills are stored in the repository's `.agents/skills` folder.

### Manage skills [#manage-skills]

Use the switch beside a skill to turn it on or off, and its menu to **Edit** or **Remove** it. **Built-in** skills come with Layerwise and cannot be edited or removed.

### Use a skill in a message [#use-a-skill-in-a-message]

Type `/` in the message box and choose the skill under **Skills**. It is added to the message as `$skill-name`, and the agent follows it for that request. The agent can also use a skill that is turned on without being asked, when your request matches its description.

## MCP servers [#mcp-servers]

### Add a server [#add-a-server]

1. Choose **Add server**.
2. Enter a **Server ID**, such as `figma`. It can contain letters, numbers, `_`, and `-`, and cannot be changed later.
3. Choose a **Connection type**:
   * **Streamable HTTP** — Enter the **Server URL** and choose **Authentication**: **None**, **API token**, or **OAuth**.
   * **Local command (stdio)** — Enter the **Command**, such as `npx`, and its **Arguments (JSON array)**. Add environment variables under **Advanced**.
4. Save the server.
5. For OAuth, choose **Authorize** on the server and sign in from your browser.
6. Turn on the server's switch. New project servers stay off until you turn them on.

Local commands run on your computer, so add only commands you trust. Tokens and other credentials are stored encrypted on this device, not in the repository.

### Check a server [#check-a-server]

Choose **Test connection** to connect to the server and list the tools it offers. The server's status shows **Connected**, **Needs authorization**, **Not tested**, **Error**, or **Disabled**.

### MCP tools in the conversation [#mcp-tools-in-the-conversation]

The conversation lists each MCP tool the agent calls. When a server needs information from you, the agent shows **MCP input required**; choose **Accept**, **Decline**, or **Cancel**.