Everything MCP Server

What this server does

Everything is not a useful server. It is a reference test fixture that exercises every part of the MCP protocol so client developers can verify their implementation. Prompts, tools, resources, sampling, progress reporting, cancellation, all wired up in one binary. Tools include trivial demos (`echo`, `add`) and protocol-exercising ones. `long_running_operation` emits progress events you can use to verify your client's progress UI and cancellation path. `sample_llm` triggers server-initiated sampling, which is one of the harder MCP features to test. Resources, prompts, and roots are also registered. Three transports ship in the same package: stdio (default), the deprecated HTTP+SSE, and the current Streamable HTTP. After `npm install -g @modelcontextprotocol/server-everything`, run `npx @modelcontextprotocol/server-everything <transport>` to pick. The full list of registered primitives lives in `docs/features.md` inside the upstream repo. When to reach for it: writing an MCP client, debugging why a feature does not work in your host, or learning what the protocol can actually do beyond simple tool calls. When not to: any real workflow. The tools are deliberately useless, and pointing a model at it will burn tokens calling `add(1, 2)` and `echo` for no benefit.

Tools included

• echo

  Echo back the input message.

• get_sum

  Add two numbers and return the sum.

• get_tiny_image

  Return a small base64-encoded image resource.

• trigger_long_running_operation

  Demonstrates progress reporting and cancellation.

• trigger_sampling_request

  Demonstrates server-initiated sampling.

The authoritative list comes from the server's tools/list at runtime. This catalog entry may lag the upstream until the next ToolPiper release.

Install in ToolPiper

ToolPiper handles the JSON configuration. Click Install, fill in any required tokens, and the server is ready in your chat.

Install in another client

{
  "mcpServers": {
    "everything": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-everything"
      ]
    }
  }
}

Example prompts

Security notes

No credentials, no network calls, no file access. Safe to run anywhere. The only practical risk is billing noise: `long_running_operation` exists specifically to emit progress events, and `sample_llm` calls back into the host's LLM. If you wire this server into an MCP host that bills per token, a curious model can rack up calls by playing with the sandbox. Not a security risk, just a budget one. Run it locally for client development and disconnect when you are done.

Related tools