Agent Plugins

Concept

The problem: context overload

Without framework context, agents guess. Your agent can call Stripe APIs, but does it know to use idempotent keys? It can query Neon, but does it understand connection pooling for serverless? Without built-in expertise, you're both manually reading documentation and refining approaches until the output is right. Agent plugins give your agent (and by extension, you) instant access to specialised knowledge, so you can work in unfamiliar domains faster.

With too much context, agents slow down. Connect five MCP servers and your agent loads 100+ tool definitions before writing a single line of code. Five servers might consume 50,000+ tokens, 40% of your context window, before your first prompt. More tools should mean better results, but unstructured context overwhelms the agent, leading to slower responses and lower quality output.

How agent plugins work

Instead of loading all MCP tools at once, agent plugins activate dynamically based on keywords in your conversation.

When you start a task, Mind:

1. Reads the task description
2. Evaluates installed agent plugins against the task
3. Loads only relevant agent plugins into context

What's in an agent plugin?

Agent plugins follow the Agent Plugins specification — an open, vendor-neutral format for packaging reusable components that extend AI agents. An agent plugin is a directory with a required manifest and optional components:

1. plugin.json — The manifest that identifies the power and declares its keywords for activation
2. skills/ — Agent Skills that provide task-specific instructions, scripts, and reference material
3. mcp.json — MCP server configuration for tool integrations (optional)
4. dev.mind/ — Mind-specific extensions like steering files (optional)

my-plugin/
├── plugin.json          # Required manifest
├── skills/              # Agent Skills
│   └── setup/
│       ├── SKILL.md
│       └── references/
└── mcp.json             # MCP server configuration

Install a Stripe plugin with one click. When you mention "payment" or "checkout," the agent plugin activates, loading Stripe's MCP tools and skills into context. When you're done with payments and move to database work, a Supabase plugin activates and Stripe deactivates.

What makes agent plugins different

• Portable plugin format — Agent plugins follow the Agent Plugins specification, an open standard supported across compatible agent clients.
• Dynamic MCP tool loading — Traditional MCP servers load all tools upfront. Agent plugins load tools on-demand, reducing baseline context usage while giving your agent access to dozens of technologies.
• Open ecosystem — Install curated agent plugins, community-built plugins from GitHub repositories, or create and share your own.
• One-click install — Browse agent plugins directly in Mind. Click "Install" and the agent plugin registers automatically.

How agent plugins differ from skills and steering

• Agent Plugins are plugins that bundle MCP tools, skills, and knowledge into a single installable package. They activate dynamically based on context. Use for integrations where you need both tools and guidance. Agent plugins can contain skills as components.
• Skills are standalone portable instruction packages that guide agents through specific tasks. They can exist independently or be bundled inside an agent plugin. Use for reusable workflows you want to share or import.
• Steering is Mind-specific context that shapes agent behaviour. It supports always, auto, fileMatch, and manual modes. Use for project standards and conventions.