How to Find MCP Servers From Inside Claude Code

To find MCP servers from inside Claude Code, install the Remote OpenClaw MCP server with claude mcp add remoteopenclaw -- npx -y remoteopenclaw, then ask your agent something like "find me an MCP server for postgres." It searches the Remote OpenClaw directory of 13,870+ MCP servers, 4,384+ agent skills, and thousands of plugins, and returns names, descriptions, links, and ready-to-paste install commands ranked by stars and installs — no browser needed.

What the Remote OpenClaw MCP server is

The Remote OpenClaw MCP server is an open-source npm package (remoteopenclaw) that turns the Remote OpenClaw directory into a tool your coding agent can call directly. Instead of opening a browser to look up an integration, your agent searches the directory in-conversation and hands back install commands you can run immediately.

It is built on the Model Context Protocol (MCP), the open standard Anthropic introduced in November 2024 to connect AI assistants to external tools and data. MCP describes itself as a "USB-C port for AI applications" in the official MCP documentation — a single standard so any compliant client can talk to any compliant server. The Remote OpenClaw MCP server is one of those servers, and the thing it serves is search over the directory itself.

Under the hood it is intentionally thin: a small wrapper over the public search API at https://www.remoteopenclaw.com/api/search. The directory currently indexes 13,870+ MCP servers, 4,384+ agent skills, and thousands of plugins for Claude Code, OpenClaw, Hermes Agent, and Codex.

The three tools it exposes

The Remote OpenClaw MCP server exposes exactly three tools, one per directory category. Each returns names, descriptions, links, and install commands, ranked by stars and installs so the most-used result lands first.

Because the agent gets the install command back as part of the result, the loop from "I need X" to "X is installed" stays inside one conversation. For deeper context on the categories, see the MCP servers hub and the agent skills hub.

How to install it in Claude Code

Installing the Remote OpenClaw MCP server in Claude Code takes a single command. Claude Code registers MCP servers with claude mcp add <name> -- <command>, where -- separates Claude's flags from the server command, per the Claude Code MCP docs.

claude mcp add remoteopenclaw -- npx -y remoteopenclaw

The -y flag tells npx to accept the package install prompt automatically, so the server does not hang on first run. After it registers, run claude mcp list to confirm the server shows up, then start asking your agent to search the directory.

Adding it to OpenClaw, Hermes, Codex, and Cursor

Any MCP-compatible client can run the Remote OpenClaw MCP server using the same npm package. For clients that use a JSON config file — OpenClaw, Hermes Agent, Codex, and Cursor — add this block to the mcpServers section:

{
  "mcpServers": {
    "remoteopenclaw": {
      "command": "npx",
      "args": ["-y", "remoteopenclaw"]
    }
  }
}

Because MCP is a shared standard now adopted by OpenAI and Google DeepMind as well as Anthropic, the same server works across editors without per-client rewrites. That is the whole point of the protocol — write the server once, call it from any compliant agent.

What searching from the editor looks like

Searching the directory from the editor replaces a browser tab with a sentence to your agent. A developer working in Claude Code asks "find me an MCP server for postgres," the agent calls search_mcp_servers, and the response comes back with ranked matches, a short description for each, a link to the listing, and the install command.

The table below contrasts the old browser flow with the in-editor flow.

For a primer on what MCP servers do once installed, the MCP servers explained guide walks through tools, resources, and how clients connect.

Limitations and Tradeoffs

The Remote OpenClaw MCP server only does search — it finds and returns install commands, but it does not run them for you. You still review and run each install command yourself, which is the right default for anything that adds tools or network access to your agent. Treat unfamiliar servers with the same caution you would any dependency.

It also depends on network access to the public search API, so it will not work fully offline, and results are only as current as the directory index. If you already know the exact package you want, calling claude mcp add directly is faster than searching. The value shows up when you do not know what exists yet, or when you want your agent to discover and compare options without leaving the editor.

Related Guides

• OpenClaw MCP Servers Explained
• How to Build Your Own MCP Server
• MCP Servers vs API Integrations: When to Use Each
• How to Find the Right Skill for Your Project