mcp-terraform-ls

An MCP (Model Context Protocol) server that wraps terraform-ls, exposing Terraform language intelligence as MCP tools. This enables AI assistants like Claude Code to understand and navigate Terraform codebases.

Prerequisites

• Node.js >= 20

• terraform-ls installed

Install terraform-ls:

# macOS
brew install hashicorp/tap/terraform-ls

# Or download from https://github.com/hashicorp/terraform-ls/releases

Related MCP server: MCP SysOperator

Installation

From GitHub Releases

Download the latest .tgz package from the releases page and install it globally:

npm install -g https://github.com/alexsavio/mcp-terraform-ls/releases/latest/download/mcp-terraform-ls-<version>.tgz

Or download the tarball first and install locally:

# Download
curl -LO https://github.com/alexsavio/mcp-terraform-ls/releases/latest/download/mcp-terraform-ls-<version>.tgz

# Install globally
npm install -g mcp-terraform-ls-<version>.tgz

Replace <version> with the actual version number (e.g., 2026.3.0).

From source

git clone https://github.com/alexsavio/mcp-terraform-ls.git
cd mcp-terraform-ls
npm install
npm run build
npm link

Usage with Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "terraform": {
      "command": "npx",
      "args": ["-y", "mcp-terraform-ls"]
    }
  }
}

If you installed from a GitHub Release, you can reference the binary directly:

{
  "mcpServers": {
    "terraform": {
      "command": "mcp-terraform-ls"
    }
  }
}

Optionally set the TERRAFORM_LS_PATH environment variable if terraform-ls is not on your PATH.

Available Tools

Tool Parameters

Position-based tools (hover, definition, references, completion):

• file — absolute path to the .tf file

• line — line number (1-based)

• character — column number (1-based)

File-based tools (diagnostics, document_symbols, format):

• file — absolute path to the .tf file

How It Works

This server acts as a bridge between MCP (used by AI assistants) and LSP (used by terraform-ls):

1. Receives MCP tool calls from the AI assistant

2. Manages a terraform-ls subprocess via stdio

3. Translates MCP requests into LSP JSON-RPC calls

4. Returns formatted results back through MCP

The LSP client lazily initializes terraform-ls on the first tool call, using the file's directory as the workspace root.

Development

npm install
npm run build
npm run dev  # watch mode

License

MIT