Agent extensions should not have to be rebuilt for every client.

LatchLoop now supports Agent Plugins, the open, vendor-neutral format for packaging reusable agent components. Any plugin that follows the specification can bring its portable components into LatchLoop, whether it comes from the LatchLoop marketplace, the wider ecosystem, or your own team.

Package skills and tools once

The Agent Plugins 1.0 specification defines a small, predictable package structure. A plugin starts with a plugin.json manifest and can include:

• Agent Skills: reusable instructions, references, assets, and scripts that teach an agent how to carry out a workflow
• MCP servers: tools and data connections described in mcp.json
• Client extensions: optional namespaced features that do not change the portable core

A small plugin can be as straightforward as this:

my-plugin/
├── plugin.json
├── skills/
│   └── summarize/
│       └── SKILL.md
└── mcp.json

With this shared structure, builders can package portable skills and MCP integrations once instead of rearranging the same pieces for every compatible client. LatchLoop users can install those plugins without waiting for each integration to be rebuilt in a proprietary format.

Add the plugins that fit your work

You can install a plugin from the LatchLoop marketplace or add a conforming Agent Plugin privately. Once installed, its skills appear alongside your other reusable skills. Its MCP tools can connect agents to the services they need.

A research plugin might pair source-finding tools with a briefing skill. A product plugin could help an agent interpret analytics before updating a roadmap. An internal plugin could package your team’s operating playbook for reuse across projects.

The standard defines the portable package. LatchLoop handles discovery, installation, authentication, project context, and tool approvals. Client-specific extensions only work in clients that support them, while standard skills and MCP configuration travel together.

Approvals stay in your control

Plugins can expose low-risk actions or tools that touch production systems, send messages, and update records. Those actions should not all run under the same rules.

LatchLoop manages approvals per project. You can require review for risky actions, turn on auto-approval for trusted workflows, and allow individual plugins or tools that would otherwise pause for approval.

Each project can use the controls that fit its work:

• stricter review for sensitive projects or new plugins
• faster execution for trusted internal workflows
• custom approval and tool availability settings based on the team, project, and system involved

Agent Plugins makes the package portable. LatchLoop controls how its capabilities run in each project.

A more portable agent ecosystem

We want people to choose agent extensions because they are useful, rather than because one client accepts their format. Builders get one interoperability target, and LatchLoop users get a wider choice of skills and tools.

Explore plugins in the LatchLoop marketplace. To build your own, read the Agent Plugins overview and versioned specification.