What xmpuspus/cloudwright does
The xmpuspus/cloudwright MCP server exposes Cloudwright’s architecture workflow to MCP-compatible coding agents. It works across AWS, GCP, Azure, and Databricks, covering 114 service keys according to the project documentation. An agent can turn a natural-language requirement into a typed YAML specification, then use that specification for cost analysis, compliance checks, infrastructure export, review, and deployment-readiness validation.
The xmpuspus/cloudwright MCP server also supports migration planning. A source estate and target environment can be represented as an ordered set of migration waves with dependencies, costs, and evidence gates. The migration model covers infrastructure, applications, data, platforms, networks, facilities, and business services across on-premises, cloud, hybrid, cross-cloud, data-center, and application moves.
How it works
The specification is the central artifact. Cost output is broken down by component and region, with a confidence flag on each line. Compliance findings map to control identifiers for frameworks including HIPAA, SOC 2, PCI-DSS, FedRAMP, GDPR, ISO 27001, and NIST 800-53. Reviews combine scoring, linting, and validation into a severity-ranked report.
Architecture exports can target Terraform, OpenTofu, Pulumi in TypeScript or Python, and CloudFormation. The project describes safe defaults such as S3 public-access blocking, encryption, and versioning; encrypted, multi-AZ RDS with deletion protection; and IMDSv2 for EC2. The plan workflow validates generated Terraform but does not apply infrastructure.
Most commands run without a model or network connection. The design, modify, chat, and ADR operations are the stated exceptions and can use either Anthropic or OpenAI credentials. MCP clients use the same engine for migration planning through the plan_migration and verify_migration tools.
Setup and configuration
The project provides a Python package named cloudwright-ai and documents installation with its CLI extra:
pip install 'cloudwright-ai[cli]'
Set ANTHROPIC_API_KEY or OPENAI_API_KEY when using the model-backed commands. Other documented commands, including cost, compliance, export, plan, review, and migration workflows, can operate offline. The CLI accepts --json for machine-readable output and --stream for streamed model output.
The xmpuspus/cloudwright MCP server is intended to be configured through the project’s integration command rather than by hand. The documented harnesses include Claude Code, Cursor, Cline, Windsurf, GitHub Copilot, Zed, Codex CLI, Junie, Kiro, and Antigravity. Aider receives a CLI-pipe recipe because it does not use MCP.
Tools and capabilities
The README groups 24 MCP tools into design, cost, validate, analyze, export, session, review, compliance, plan, and migration areas. Supported capabilities include:
- Generate and modify architecture specifications.
- Estimate regional and component-level costs.
- Check compliance controls and produce traceability output.
- Lint, score, and review specifications.
- Analyze blast radius and single points of failure.
- Scan specifications and exported HCL for security issues.
- Enforce nine built-in policy-as-code rules.
- Compare a design with Terraform state and optionally produce remediation output.
- Export infrastructure and render ASCII, Mermaid, D2, or web-canvas diagrams.
- Plan and verify migrations using evidence files.
Limitations and notes
The xmpuspus/cloudwright MCP server plans and checks migrations; it does not copy data, apply infrastructure, switch traffic, or execute a cutover. A migration verification can remain closed when blocking evidence is missing, and the command returns exit code 2 in that case.
Cloudwright’s compliance checks occur before resources exist, based on the specification and generated artifacts. Checkov can be included when it is available on the system PATH. The plan command validates Terraform output but never applies it. Canvas edits are deterministic frontend changes and do not invoke a model.