terrashark MCP — 使用場景, 安裝 & 即時演示

為什麼要用

核心特性

Grounded in HashiCorp Terraform best practices
Eliminates hallucinated resource arguments
Modular structure by default (modules/, envs/)
Security-first defaults (least privilege, encryption on)
State layout guidance (remote backend, workspace patterns)
Works with Claude Code and Codex

即時演示

實際使用效果

terrashark-skill.replay ▶ 就緒

0/0

安裝

選擇你的客戶端

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "terrashark-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/LukasNiessen/terrashark",
        "~/.claude/skills/terrashark"
      ],
      "_inferred": true
    }
  }
}

開啟 Claude Desktop → Settings → Developer → Edit Config。儲存後重啟應用。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "terrashark-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/LukasNiessen/terrashark",
        "~/.claude/skills/terrashark"
      ],
      "_inferred": true
    }
  }
}

Cursor 使用與 Claude Desktop 相同的 mcpServers 格式。專案級設定優先於全域。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "terrashark-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/LukasNiessen/terrashark",
        "~/.claude/skills/terrashark"
      ],
      "_inferred": true
    }
  }
}

點擊 Cline 側欄中的 MCP Servers 圖示，然後選 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "terrashark-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/LukasNiessen/terrashark",
        "~/.claude/skills/terrashark"
      ],
      "_inferred": true
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "terrashark-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/LukasNiessen/terrashark",
        "~/.claude/skills/terrashark"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列，而非映射。

~/.config/zed/settings.json

{
  "context_servers": {
    "terrashark-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/LukasNiessen/terrashark",
          "~/.claude/skills/terrashark"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add terrashark-skill -- git clone https://github.com/LukasNiessen/terrashark ~/.claude/skills/terrashark

一行命令搞定。用 claude mcp list 驗證，claude mcp remove 移除。

使用場景

實戰用法： terrashark

How to scaffold a production-grade AWS VPC module

👤 Platform engineers starting a new environment ⏱ ~45 min intermediate

何時使用： Day one of a new AWS account; you want a VPC module that won't embarrass you in 6 months.

前置條件

Skill installed — git clone https://github.com/LukasNiessen/terrashark ~/.claude/skills/terrashark
terraform CLI 1.6+ — brew install terraform

步驟

Generate module

Use terrashark. AWS VPC module, 3 AZs, public + private + db subnets, NAT gateways per AZ, flow logs on.✓ 已複製

→ Module with variables.tf, main.tf, outputs.tf — no hallucinated args
Add security defaults

Ensure default security group denies all, flow logs go to CloudWatch with KMS-CMK encryption.✓ 已複製

→ Security group + logging + KMS wired
Validate

Run terraform init + validate + plan with example tfvars.✓ 已複製

→ Clean plan output, no errors

結果： A VPC module that terraform validates clean and follows AWS best practices.

注意事項

LLM suggests deprecated aws_vpc_endpoint arg — TerraShark grounds in current docs; if drift still happens, ask it to cite the HashiCorp page
Module too monolithic — Ask for split: networking module vs. endpoints module separately

搭配使用： aws-skill

Fix LLM-generated Terraform that doesn't plan

👤 Engineers cleaning up AI-generated IaC ⏱ ~20 min intermediate

何時使用： Another LLM gave you Terraform that looks reasonable but terraform plan fails on unknown arguments.

步驟

Point at the broken file

terraform plan fails on modules/rds. Use terrashark to fix each invalid argument with the current correct form.✓ 已複製

→ Per-error fix with HashiCorp doc reference
Validate

Re-run plan; if any errors remain, iterate.✓ 已複製

→ Clean plan

結果： A file that plans clean.

Design a remote state layout for multi-env

👤 Teams graduating from local state ⏱ ~60 min advanced

何時使用： You have dev + staging + prod and one giant terraform.tfstate.

步驟

Propose layout

Design S3 remote backend with per-env state files and DynamoDB locking. 3 envs.✓ 已複製

→ Backend config per env with key separation + locking
Migration plan

Plan the migration from a single local state to this layout, step by step.✓ 已複製

→ terraform state mv steps + safety checkpoints

結果： A state layout you can migrate to without losing state.

注意事項

Accidentally deleting state during migration — Always pull and back up state files before any mv operation

組合

與其他 MCP 搭配，撬動十倍槓桿

terrashark-skill + aws-skill

AWS skill sizes architecture, TerraShark implements IaC

Design the architecture with aws-skill, implement as Terraform with terrashark.✓ 已複製

terrashark-skill + claude-code-owasp-skill

Security-review the generated IaC

After terrashark scaffolds the module, run OWASP security review on the IAM and networking blocks.✓ 已複製

工具

此 MCP 暴露的能力

工具	輸入參數	何時呼叫	成本
Module scaffolding	provider, resource shape	Greenfield IaC	Claude tokens
Hallucination fix	tf file + error log	Cleaning up AI-generated TF	Claude tokens
State layout design	envs, team shape	Multi-env setups	Claude tokens
Security defaults	-	Pre-prod review	Claude tokens

成本與限制

運行它的成本

API 配額: None
每次呼叫 Token 數: 10-30k per module
費用: Free
提示: Generate module by module, not whole environment in one shot.

安全

權限、密鑰、影響範圍

憑證儲存： No credentials. TerraShark doesn't run terraform apply itself.

資料出站： Prompts go to Claude. Terraform plan/apply hits your cloud provider using your local creds.

Always review generated IAM policies before apply. LLMs can silently over-grant.

故障排查

常見錯誤與修復

terraform validate fails on unknown argument

Ask TerraShark to reconcile against the current provider doc URL; provider version may be newer than the skill's prompt context.

驗證： terraform providers

Plan creates unwanted destroys

Usually state-drift or implicit renaming. Run terraform plan -target to isolate before apply.

替代方案

terrashark 對比其他方案

替代方案	何時用它替代	權衡
Hand-writing with official docs open	You know the provider deeply	Slower

terrashark

為什麼要用

核心特性

即時演示

實際使用效果

安裝

選擇你的客戶端

使用場景

實戰用法： terrashark

How to scaffold a production-grade AWS VPC module

前置條件

步驟

注意事項

Fix LLM-generated Terraform that doesn't plan

步驟

Design a remote state layout for multi-env

步驟

注意事項

組合

與其他 MCP 搭配，撬動十倍槓桿

工具

此 MCP 暴露的能力

成本與限制

運行它的成本

安全

權限、密鑰、影響範圍

故障排查

常見錯誤與修復

替代方案

terrashark 對比其他方案

更多

資源