Plugin Builder

Cursor Plugin Builder

Build Cursor plugin scaffolds. Optionally integrates with Runlayer MCP for tool discovery.

Before You Start: Check Runlayer MCP (Optional)

1. Check if Runlayer MCP tools are available (e.g., mcp__runlayer__list_servers)
2. If available: call mcp__runlayer__list_servers(scope="accessible") to discover available integrations
3. If NOT available: proceed without Runlayer integration. The plugin can still include rules, skills, hooks, agents, and commands.

Directory Context

1. Run ls -la to understand the current directory structure
2. Check if this is already a plugin marketplace (look for .cursor-plugin/marketplace.json)
3. If it's a marketplace, new plugins should be created as subdirectories
4. Always confirm the target location with the user before creating files

Workflow

Step 1: Gather Requirements

Use AskQuestion to collect structured input:

Question 1: Plugin basics

• Plugin name (kebab-case slug, e.g., customer-support)
• 1-2 sentence description

Question 2: Components to include (multi-select)

• Rules (.mdc files) -- persistent AI guidance
• Skills (SKILL.md) -- specialized agent capabilities
• Commands (.md) -- explicit user-invoked actions
• Hooks (hooks.json) -- automation on agent events
• Agents (.md) -- custom agent behaviors
• MCP servers (.mcp.json) -- external tool integrations

Question 3: If skills selected -- automation preference

• "Fully automated" -- Skills trigger automatically based on context
• "Manual triggers only" -- Commands require explicit invocation
• "Both" -- Skills for automation, Commands as explicit fallback

Question 4: If MCP tools needed -- tool categories (from Runlayer if available)

• Present discovered servers grouped by category (CRM, ticketing, analytics, etc.)
• If no Runlayer, ask which external services to integrate

Step 2: Generate Plugin Structure

<plugin-slug>/
├── .cursor-plugin/
│   └── plugin.json
├── rules/                    # if rules selected
│   └── <rule-name>.mdc
├── skills/                   # if skills selected
│   └── <skill-name>/
│       └── SKILL.md
├── commands/                 # if commands selected
│   └── <command-name>.md
├── agents/                   # if agents selected
│   └── <agent-name>.md
├── hooks/                    # if hooks selected
│   └── hooks.json
├── .mcp.json                 # if MCP tools needed
└── README.md

File Templates

plugin.json:

{
  "name": "<plugin-slug>",
  "description": "<1-2 sentence description>",
  "version": "1.0.0",
  "author": { "name": "<author>" },
  "keywords": ["<relevant>", "<tags>"]
}

Rule (.mdc):

---
description: <What this rule enforces>
alwaysApply: true
---

## <Rule Title>

- Rule point 1
- Rule point 2

For glob-scoped rules (apply only to matching files):

---
description: <What this rule enforces>
globs:
  - "**/*.ts"
  - "**/*.tsx"
---

SKILL.md:

---
name: <skill-name>
description: <When this skill should activate. Be specific about triggers and contexts.>
---

# <Skill Title>

## When to Use
<Specific triggers and contexts>

## Workflow
<Step-by-step guidance>

## Best Practices
<Domain-specific guidance>

Command (.md):

---
name: <command-name>
description: <What this command does>
---

# <Command Title>

<Instructions for executing this command>

## Usage
Invoke via `/plugin-name:command-name`

## Workflow
<Step-by-step execution>

Agent (.md):

---
name: <agent-name>
description: <Brief description of the agent's purpose>
---

# <Agent Title>

<System prompt and behavioral instructions for the agent>

hooks.json:

{
  "hooks": {
    "<hookEvent>": [{ "command": "./scripts/<script>.sh" }]
  }
}

Available hook events: sessionStart, sessionEnd, preToolUse, postToolUse, beforeShellExecution, afterShellExecution, beforeMCPExecution, afterMCPExecution, beforeReadFile, afterFileEdit, beforeSubmitPrompt

.mcp.json (only if MCP tools needed):

{
  "mcpServers": {
    "<server-name>": {
      "url": "<server-url>"
    }
  }
}

If Runlayer is available, use URLs from mcp__runlayer__list_servers results. Never guess URLs.

Step 3: Generate CONNECTORS.md (if MCP tools used)

# Connectors

| Category | Available Servers |
|----------|-------------------|
| CRM | Salesforce, HubSpot |
| Ticketing | Zendesk, Jira |

## Configuration
Configure MCP servers in `.mcp.json` based on which services you use.

Plugin Examples

Security Plugin

• Rules: secrets-hygiene, code-review-standards
• Skills: vulnerability-scan, dependency-audit
• Hooks: afterFileEdit (lint), beforeShellExecution (validate)

DevOps Plugin

• Skills: incident-response, deployment-review
• Commands: /deploy-status, /incident-summary
• MCP: Monitoring (Datadog), CI/CD (GitHub Actions)

Customer Support Plugin

• Skills: ticket-triage, response-drafting
• Commands: /escalate, /draft-response
• MCP: Ticketing (Zendesk), CRM (Salesforce)

Testing

After generating, instruct the user to test locally:

1. Install the plugin from the local directory in Cursor
2. Try invoking skills by describing relevant tasks
3. Try commands via /plugin-name:command-name
4. Verify rules apply by checking agent behavior in matching contexts

Submission

When ready to publish:

1. Ensure .cursor-plugin/plugin.json manifest is valid
2. All rules/skills/agents/commands have proper frontmatter
3. Push to a public Git repository
4. Submit at cursor.com/marketplace/publish