Plugins

Plugins extend Proliferate agents with the tools, context, and capabilities they need to do their work.

A plugin bundles three things into one unit:

• Skills- procedural workflows the agent can follow.
• MCPs- Model Context Protocol servers that expose tools and resources.
• Proliferate-built capabilities- first-party features like subagent management, Computer Use, and Browser Use.

Enable a plugin and your agent gets everything inside it- one toggle, one permission scope.

Per-session scope

Plugins are scoped per session. When a session starts, Proliferate builds a plugin bundle for it: the enabled skills are mounted, the MCP servers it needs are bound according to the workspace's MCP binding policy, and the agent gets a single coherent set of capabilities for the lifetime of that session.

This means:

• The same workspace can run two sessions with different plugins enabled.
• Disabling a plugin mid-session takes effect on the next session.
• Plugin state doesn't leak across sessions.

Skills, MCPs, and credential bindings

Each plugin can declare:

• Skills- with a display name, description, instructions, and any required MCP servers.
• MCP servers- the tool/resource servers the skills depend on.
• Credential bindings- the auth tokens or API keys the MCPs need.

Skills with defaultEnabled: true are mounted automatically when a session launches. Others can be turned on per session. MCP servers only mount when they bind successfully against the workspace's binding policy.

The plugin catalog

Plugins are distributed through the Proliferate cloud catalog. The desktop app pulls the available set, you choose which to enable per workspace, and the bundle is materialized into the session at launch.

First-party Proliferate plugins include subagent management, Computer Use, and Browser Use. Community MCPs are listed in the catalog alongside them.

Trust and permissions

Plugin-scoped permission prompts let you approve or deny specific capabilities at the moment the agent tries to use them. For sensitive plugins (Computer Use, Browser Use, anything that writes to external systems) this is your last line of defense before the agent acts.

Common uses

Plugins help with:

• Reading product specs, design files, or analytics.
• Pulling issue, support, or PR context from connected systems.
• Updating external systems after review (Linear, Slack, email).
• Driving desktop apps and browsers via Computer / Browser Use.
• Spawning subagents for parallel investigation or implementation.

Enable plugins in the workspace settings. Inside a session, a plugin's tools show up alongside the agent's built-in ones.