A plugin packages the guidance, connections, and interface an AI needs to do a particular kind of work—then makes that capability discoverable and portable.
The label identifies what it is. Inside, instructions explain the work, adapters connect equipment, and a control panel may expose the result.
A plugin can bundle skills, an MCP server connection, and optional UI behind one installable identity.
The manifest is the label. It gives the package an identity and points the host to what is inside.
Not every plugin uses every layer. The smallest useful package is usually the best starting point.
Instructions coordinate a repeatable method while tools read or change external systems.
Use when existing tools are enough, but the agent needs a reliable procedure or specialist guidance.
Use when the host needs new tools or data access and the operation is already self-explanatory.
Use when a workflow must guide when and how a set of external tools should be used.
Credentials determine which account and data the tools can reach.
Tool schemas define the operations exposed to the model.
High-impact actions should stay legible and intentional.
Structured results and UI help people inspect the outcome.
Encodes the method and judgment for a repeatable workflow.
Connects the AI to tools, services, authentication, and structured results.
Bundles the capability behind an identity people can discover and install.