Turf

The infrastructure engine built for AI agents.

Turf™ is a self-hosted, autonomous infrastructure-as-code tool, built on OpenTofu™. Terraform-compatible — full HCL and module-registry support — exposed to AI agents as an infrastructure MCP server.

turf.build · Get started · Examples · Community

brew install restatedev/tap/restate-server turfbuild/tap/turf
brew services start turf                                        # the engine, now and at every login
npx add-mcp@2 "$(command -v turf-mcp-server)" --name turf -g    # connect your agent

Then, in Claude Code: /turf:guide.

This installs four binaries on your PATH: turf-mcp-server, the MCP server your agent talks to; the engine behind it, turf-engine and turf-builtin-provider, which brew services start turf runs as a service (turf-mcp-server engine status says where); and the Turf CLI, turf. Without Homebrew, curl -fsSL https://get.turf.build/install.sh | bash installs the same into ~/.turf/bin.

In Claude Code and Claude Desktop, Turf runs on your Claude subscription — no separate model API key.

The npx add-mcp@2 … --name turf -g line registers the server with every agent on this machine. It detects the ones you have installed — Claude Code, Claude Desktop, Cursor, Codex, Gemini CLI, VS Code, Windsurf, Zed and more — lets you pick from the list, and writes each config with the server's absolute path, so an app launched from the Dock finds it too. It needs Node 18+ and uses add-mcp, Neon's open-source installer; running it again replaces the entry. Restart Claude Desktop or Cursor after the config changes.

Without Node, any client with an mcp add command registers the server in one line:

claude mcp add turf --scope user -- turf-mcp-server             # Claude Code
codex  mcp add turf -- turf-mcp-server                          # Codex CLI
gemini mcp add -s user turf turf-mcp-server                     # Gemini CLI
code --add-mcp '{"name":"turf","command":"turf-mcp-server"}'    # VS Code

For a client without one, add the standard block to its MCP config, with the path command -v turf-mcp-server prints:

{ "mcpServers": { "turf": { "command": "/path/to/turf-mcp-server" } } }

claude mcp add --scope user turf -- turf-mcp-server        # every project (~/.claude.json)
claude mcp add --scope project turf -- turf-mcp-server     # this repo, committed to .mcp.json

claude mcp add --scope project --env AWS_PROFILE=turf-dev turf -- turf-mcp-server

codex mcp add turf -- turf-mcp-server
codex mcp add turf --env AWS_PROFILE=turf-dev -- turf-mcp-server

[mcp_servers.turf]
command = "turf-mcp-server"

gemini mcp add -s user turf turf-mcp-server                 # every project (~/.gemini/settings.json)
gemini mcp add -s project turf turf-mcp-server              # this repo (.gemini/settings.json)
gemini mcp add -s project -e AWS_PROFILE=turf-dev turf turf-mcp-server

// VS Code — root key is "servers", each entry needs a "type"
{ "servers": { "turf": { "type": "stdio", "command": "turf-mcp-server" } } }

// Zed — root key is "context_servers", each entry needs "source": "custom"
{ "context_servers": { "turf": { "source": "custom", "command": "turf-mcp-server" } } }

{
  "mcpServers": {
    "turf": {
      "url": "https://turf.example.com/mcp",
      "headers": { "Authorization": "Bearer ${TURF_TOKEN}" }
    }
  }
}

turf chat                          # interactive, in a terminal
cd path/to/config && turf up       # deploy or reconcile a configuration

The CLI drives the same server and engine with its own agent, so it brings its own model: set ANTHROPIC_API_KEY (or GOOGLE_API_KEY, OPENAI_API_KEY), run a local model, or point it at a private endpoint. Its README: turfbuild/turf.

Turf is built on OpenTofu and a small set of MPL-2.0 forks we maintain in the open:

• opentofu — our OpenTofu fork
• provider-client — OpenTofu provider-client fork
• terraform-provider-tfcoremock — mock provider used in tests
• docker-agent — downstream cagent fork powering the CLI

Turf is free for noncommercial use under the PolyForm Noncommercial License 1.0.0. Commercial use, including a company's internal use, needs a license: https://turf.build/licensing. The turf CLI is open source (MPL-2.0).

Turf sends anonymous usage telemetry to Turf LLC by default: the version, the platform, counts of resources, changes and runs, and an anonymous installation id. It never sends resource contents, credentials, state, variable values or names. Turn it off with turf telemetry off, or with TURF_TELEMETRY=off or DO_NOT_TRACK=1 in the engine's environment; the full field list and the endpoint are at https://turf.build/telemetry.

• Questions, ideas, show and tell: Discord.
• Bugs and feature requests: turfbuild/turf issues.
• Release notes by email: the newsletter. Follow Turf on LinkedIn and X.
• Evaluating Turf at a company? The pilot program.
• Anything else: contact@turf.build. All of it at https://turf.build/community/.

Turf has a transparent supply-chain posture: every push runs continuous scanning (govulncheck, osv-scanner, gosec, trivy), and releases ship SBOMs (SPDX + CycloneDX), OpenVEX disposition documents, and scan reports with GitHub-native build & SBOM attestations. See our security policy — report issues privately to security@turf.build.