MCP Server Browser

What agent-infra/mcp-server-browser MCP server does

The agent-infra/mcp-server-browser MCP server exposes browser automation through the Model Context Protocol and uses Puppeteer as its browser control layer. Its documented scope is browser interaction, with support for connecting to either a browser running locally or one hosted remotely.

This makes the server relevant for agents that need to work with browser-based interfaces. It can fit workflows where an agent must operate a web browser as part of a larger task, especially when the browser may run in a separate environment from the MCP client. The supplied material does not enumerate the available MCP tools or describe specific page, navigation, input, or extraction operations, so those details should be verified in the source repository before implementation.

How it works

Puppeteer provides the browser automation implementation, while MCP supplies the interface through which an agent can access the server. The browser connection can target a local browser process or a remote browser endpoint. This gives deployments a choice between keeping browser execution on the same machine as the MCP process and placing it elsewhere.

The repository location is within the UI-TARS Desktop project, whose broader documentation describes local and remote browser operators. That broader project context does not, by itself, define additional capabilities for the agent-infra/mcp-server-browser MCP server. Treat the browser server as the relevant component when evaluating compatibility and deployment requirements.

Setup and configuration

The provided excerpt does not include an installation command, package name, startup command, connection URL format, or environment variables for the agent-infra/mcp-server-browser MCP server. It also does not state which MCP clients have been tested. Consult the repository’s package files and browser-server documentation for the exact launch configuration.

A deployment must account for the browser location: local mode requires access to a local browser, while remote mode requires the server to connect to a remote browser. The excerpt does not specify the supported remote connection protocol, authentication method, browser versions, or required Puppeteer options.

Tools and capabilities

Documented capabilities are limited to:

• Automating browser activity through Puppeteer.
• Connecting to a local browser.
• Connecting to a remote browser.
• Providing those browser capabilities through MCP.

The available tool names, argument schemas, navigation behavior, page interaction coverage, and return formats are not described in the supplied material.

Limitations and notes

The agent-infra/mcp-server-browser MCP server is documented only at a high level in the supplied excerpt. There is not enough information to confirm supported browsers, concurrency behavior, session lifecycle, remote endpoint details, authentication, or security controls. Browser automation can also depend on the target site and the runtime environment, but the excerpt provides no compatibility matrix or operational guidance.

Before adopting it, inspect the server package directly for its MCP tool definitions and startup instructions. Confirm how local and remote browser connections are selected, whether a browser must be launched separately, and what isolation or access controls are available in the chosen deployment.