Install the Mélodium MCP Server

The Mélodium MCP server gives an AI assistant live access to Mélodium program validation and standard library reference lookup, on top of what the skill already teaches it. Instead of relying only on what it remembers about the language, the assistant can check a program file for errors and look up the exact signature of a treatment or function as it writes your pipeline.

It is published as the melodium-mcp crate, with source in the melodium-mcp directory of the Mélodium repository. It reuses the same loading code as melodium check and melodium info, so its answers match what the CLI would report, and it validates programs in mock mode: reading the file being checked is the only I/O it performs.

Tools It Exposes

check_program: parse and validate a Mélodium program file (.mel, Compo.toml, or .jeu), returning structured errors and the list of available entrypoints.
get_program_info: describe each entrypoint of a program file: identifier, documentation, and parameters.
list_library_elements: list standard library treatments, functions, models, contexts, and data types, optionally filtered by area (e.g. std/flow, http) and/or kind.
describe_element: full signature of one standard library element by identifier (e.g. std/flow::emit): documentation, generics, parameters, and, for treatments, inputs, outputs, required models and contexts.
search_reference: keyword search across standard library identifiers and documentation.

Install the Binary


cargo install melodium-mcp

This installs a melodium-mcp binary to your Cargo bin directory (~/.cargo/bin by default), which should already be on your PATH if you have Rust installed. To build from a clone of the repository instead:


cargo build --release --package melodium-mcp

The resulting binary is at target/release/melodium-mcp.

Claude Code


claude mcp add melodium -- melodium-mcp

Use claude mcp add melodium --scope project -- melodium-mcp to share the configuration with your team through the repository instead of keeping it local to your machine.

Claude Desktop

Add the server to your claude_desktop_config.json:


{
  "mcpServers": {
    "melodium": {
      "command": "melodium-mcp"
    }
  }
}

Cursor

Add the same server definition to .cursor/mcp.json, at the project root for a team-shared configuration or in your home directory for a personal one:


{
  "mcpServers": {
    "melodium": {
      "command": "melodium-mcp"
    }
  }
}

Zed

Add the server under context_servers in your settings.json:


{
  "context_servers": {
    "melodium": {
      "command": {
        "path": "melodium-mcp"
      }
    }
  }
}

Other Tools

Any MCP-compatible client that supports stdio servers can run melodium-mcp the same way: point its configuration at the binary path (or plain melodium-mcp if it’s on your PATH), with no arguments needed.

Going Further

With the MCP server connected, ask the AI to validate what it writes as it goes, for example “check this program with the MCP server before showing it to me” or “look up the signature of std/flow::emit”. Combine it with the skill and the pipeline-building workflow for the most reliable results.

Install the Mélodium Skill Build a Pipeline with AI