Rollup

You can run an MCP server that exposes built-in tools and a persistent HTTP transport to connect AI clients. It coordinates with your build tooling to provide contextual information, inspect configurations, and even influence the build process through MCP tools.

How to use

You integrate an MCP server into your build workflow by running your chosen MCP-enabled toolchain and connecting your MCP client (such as Cursor or an IDE plugin) to the server over HTTP or SSE. The server provides a set of built-in tools for analyzing module dependencies, inspecting build configurations, and debugging build errors. It also supports custom tools you implement via the UnpluginMcpTool interface. Once connected, the MCP client can query contextual data, request insights, and trigger actions that assist during development and debugging.

How to install

# Prerequisites
node >= 14.x
pnpm >= 7.x (recommended) or npm

# Install the MCP plugin in your project
pnpm add -D unplugin-mcp

# If you are using a bundler-specific variant, install that instead (shares codebase but exports the plugin for a specific bundler)
# Example for a Rollup setup
pnpm add -D rollup-plugin-mcp

# Note: If you want to enable built-in tools, you will typically configure them in your bundler config.

Additional setup and usage notes

You can enable built-in MCP tools by registering them in your plugin options. For example, you can expose modules that analyze dependencies, inspect build configuration, and debug errors. The server runs in the background and remains active in watch mode, allowing continuous AI interaction as you develop.

Connection to an MCP client

To connect an MCP client, add a server entry in your client’s configuration that points to the MCP HTTP stream exposed by the server. The example below shows a typical setup where the client connects to a local server in SSE mode.

{
  "mcpServers": {
    "rollup": {
      "url": "http://localhost:14514/mcp/sse",
      "type": "http",
      "args": []
    }
  }
}

Available tools

ModuleTool

Analyzes module dependencies and imports to help understand how your code is wired together within the build.

BuildConfigTool

Inspects and exposes the current build configuration used by your toolchain, enabling AI-assisted validation and adjustments.

BuildErrorTool

Helps diagnose and debug build errors by providing contextual information about failures and warnings.

BundleSizeTool

Inspects the size and composition of the bundle to aid in optimization and performance reviews.