Get the FREE Ultimate OpenClaw Setup Guide →

mcpx

Turn MCP servers into composable CLIs.

Installation
Run this command in your terminal to add the MCP server to Claude Code.
Run in terminal:
Command
claude mcp add --transport stdio lydakis-mcpx npx -y @agentdeskai/browser-tools-mcp@1.1.0

How to use

mcpx turns MCP servers into composable CLIs. You can list available servers, view the tools each server exposes, and call a tool directly. Tools are invoked by specifying the server name followed by the tool you want to run, for example: mcpx <server> <tool>. Output from tools is passed through unchanged, so you can pipe or JSON-parse it as needed. The README’s example configurations show how you can auto-discover servers or point mcpx at a direct MCP endpoint and start calling tools immediately.

The Browser Tools MCP server (example configured as browser-tools) exposes a set of browser-related utilities. With mcpx you can run a tool from that server like: mcpx browser-tools open-url --url https://example.com or mcpx browser-tools list-caches, depending on what the server exposes. Each tool can also provide a built-in --help output with its schema, thanks to mcpx’s schema-aware help. If you enable JSON output, you’ll get machine-readable responses that you can further process or feed into other tools.

You can install a server config automatically with mcpx add or use an ephemeral source by calling mcpx <source> directly. Shims and server-specific skills can be generated for tighter integration, and you can cache results per server to improve performance during repeated calls.

How to install

Prerequisites:

  • A system with Go installed (for building mcpx) or use prebuilt binaries if provided.
  • Optional: npm or Python package managers if you prefer alternative installation methods mentioned in the docs.

Install from source or use a prebuilt binary:

  1. Go-based build (from repository root): go build ./... ./mcpx --version

  2. If you prefer the npm-based distribution (as documented in the reference): npm install -g mcpx-go

  3. Alternative Python/PyPI path (for mcpx-go): pip install mcpx-go

  4. If you’re going to rely on auto-discovery, ensure you have network access to MCP endpoints and that mcpx can reach them. To use a direct MCP endpoint, you can run: mcpx https://docs.mcp.cloudflare.com/mcp

  5. Verify installation by listing available servers: mcpx # list servers mcpx <server> # list tools

Prerequisites summary: ensure you have the runtime(s) supported by your chosen installation path (Go toolchain or npm/pip), and access to MCP endpoints you intend to manage.

Additional notes

Tips and common notes:

  • Ephemeral sources: You can run tools directly from a source without adding it to your config; these runs last for the daemon’s lifetime.
  • Caching: Use --cache=<duration> to cache tool responses or --no-cache to bypass. Set per-server defaults with default_cache_ttl in the server config.
  • Output modes: Use --json for machine-readable output and -v to include origin metadata. These can be combined for structured logs.
  • Server discovery: mcpx can auto-discover existing MCP server configs or you can bootstrap a config with mcpx add. Server entries typically live in ~/.config/mcpx/config.toml.
  • Shims: You can install local shims so a server name behaves like a standalone command, e.g., mcpx shim install github, which creates a wrapper script in your PATH.
  • Server-specific skills: You can generate server-specific guidance or instruction sets to tailor usage per server.
  • If a server requires authentication or tokens (e.g., private endpoints), you may pass credentials via environment variables or headers as part of the server config.

Related MCP Servers

OpenClaw setup in under 5 minutes

Deploy your agent in 1 click. Use KILOPARTNERSMAY for 50% off your first month.