MCP

Jugg MCP is a local JSON-RPC service that exposes plugin features to agents. It uses the same underlying features as the CLI. The CLI wraps MCP in a command-line interface and is usually better suited to everyday agent use.

TIP

If you only need an agent to compile, deploy, run tests, or operate a device, install the Jugg CLI Skill first. Use this page only when your client must configure an MCP server directly.

Service information

Item	Value
Port range	`12320..12329`
Path	`/jugg-mcp`
Protocol	HTTP + JSON-RPC `2.0`
Global tools	`version`, `list-projects`
Non-global tools	All require `projectDir`

When multiple Android Studio instances are open, ports increment within the range. A client should discover the port first or use the CLI's port discovery.

Call model

The business result returned by an MCP tool is in structuredContent:

json

{
  "status": "OK",
  "message": "...",
  "data": {},
  "artifacts": [],
  "errorCode": null
}

Protocol success does not mean business success. The client must read structuredContent.status and, for compilation and deployment tools, continue reading fields such as isCompileSuccess and isDeploySuccess.

Public tools

Category	Tools
Projects and version	`version`, `list-projects`
Build and deployment	`compile`, `deploy`, `gradle-build`, `clean-reinstall`, `get-compile-status`
Testing	`instrument`
Runtime	`restart`, `wait-logs`, `activity-stack`
Devices	`devices`
UI	`layout-dump`, `view-locate`, `view-inspect`, `tap`
Remote diagnostics	`ssh-info`
Status	`status`

layout-verify and figma-layout-verify are not currently public tools unless they are registered in the public tool list later.

Asynchronous compilation

compile, deploy, gradle-build, and instrument may return a running state:

text

compile/deploy/gradle-build/instrument
  -> Return jobId
  -> Call get-compile-status(projectDir, jobId)
  -> Continue until success / failed / canceled

The CLI includes polling. When using MCP directly, poll get-compile-status yourself. You can also pass waitTimeoutMs to reduce empty polling.

Why the CLI is usually preferred

Compared with configuring MCP directly, the CLI offers these advantages:

The CLI is packaged with the skill, making correct instructions easier for an agent to obtain.
The CLI can combine commands, scripts, and pipelines, which suits continuous terminal and agent workflows.
The CLI includes asynchronous polling and heartbeats, reducing agent-side polling overhead.
CLI output modes support people, agents, and scripts.
When both CLI and MCP are available, agent instructions should define one preferred entry point to avoid mixing both interfaces in the same task.

When to use MCP directly

MCP is suitable when:

Your platform can configure only MCP servers.
You need to manage every tool through one MCP client.
You need to read the MCP tools/list schema directly.

In other situations, prefer the CLI:

Normal terminal use.
Compilation verification by an agent after code changes.
Long-running tasks that require frequent polling when the client has no effective waiting mechanism.

MCP ​

Service information ​

Call model ​

Public tools ​

Asynchronous compilation ​

Why the CLI is usually preferred ​

When to use MCP directly ​

Related pages ​