MCP / Work from your editor
MCP config
Connect the AI you already use. It checks your application and your infrastructure, then tells you what to change. Add SkaleAgents to your client, then approve access in your browser. Your client handles credentials automatically.
One URL / Connect with OAuth
Choose your MCP client
Connect your assistant to the hosted SkaleAgents MCP server. Sign in with Google and approve access through OAuth. No API keys or local server to install.
Every account connects one AI client. Paid plans connect more. See plans
- 1. Add SkaleAgents
Choose your client below and add the server URL using its setup steps.
- 2. Connect with OAuth
Connect in your client, sign in with Google if needed, and approve MCP access.
- 3. Review the codebase
Ask your editor to scan Terraform, CloudFormation, Kubernetes, Pulumi YAML, Pulumi programs, or Docker Compose with scan_iac. The scan looks for a public database, wildcard permissions, a Docker socket, and a missing memory limit. For other source, use review_architecture, which includes a disabled timeout and a one-connection pool. You can also paste one file into a web audit.
Cursor setup
Save this in .cursor/mcp.json at your project root. Open Cursor Settings, then Tools & MCP. Enable skaleagents and use Connect to complete OAuth.
{
"mcpServers": {
"skaleagents": {
"url": "https://skaleagents.com/mcp"
}
}
}What you get back
Sign in with Google and click Allow. A review of this public database sample names the finding and the exact line change.
resource "aws_db_instance" "app" {
publicly_accessible = true
}
Database publicly accessible. publicly_accessible = true becomes publicly_accessible = false.
Need to reconnect?
Use Connect or Authenticate in your client's MCP panel to start OAuth again. For Codex, run codex mcp login skaleagents. If the client timed out while you signed in, retry the connection.
Your client stores OAuth credentials and refreshes access automatically. Disconnect it below to revoke access. Signing out of this website does not disconnect your MCP clients.
Review the whole application
Give your connected assistant a review brief. It reads the relevant files, uses SkaleAgents to check the structure, and asks about the decisions behind your design.
Optional. Add the important user flows or a design decision you want challenged.
Act as an independent architecture consultant using SkaleAgents. Does this application's architecture fit its purpose and constraints? Review routing and rendering, module boundaries, data access, authorization, failure handling, testing, and deployment. Start with plan_architecture_review using this repository's file inventory. Read the relevant files, including package.json and tsconfig.json, then call review_application_architecture. Ask me up to three follow-up questions at a time and carry my answers forward in context. Explain what is sound, what needs changes, and what needs more evidence, with file references and practical alternatives.
Paste this into OpenCode or another connected assistant while the project is open. SkaleAgents returns source-backed checks and review questions; your assistant reasons through the design tradeoffs with you.
plan_architecture_review
Start a consultation with a file inventory, project goals, and a focused set of follow-up questions.
review_application_architecture
Map related source files and check Next.js boundaries. Review the whole design with evidence and questions across data, authorization, reliability, and delivery.
review_architecture
Review an individual source snippet or infrastructure configuration for security, reliability, and cost concerns, including a disabled timeout, a one-connection pool, and an unbounded retry limit.
scan_iac
Scan Terraform, CloudFormation, Kubernetes, Pulumi YAML, Pulumi programs, or Docker Compose for a public database, wildcard permissions, a Docker socket, a missing memory limit, unencrypted storage, and host access.
Scans read the file text your editor sends. They do not connect to a live cloud account. Pulumi programs are scanned as text, so computed values and external modules are not evaluated. Results include coverage limits, and runtime behavior is not inspected.
Manage your connections
See which clients have access and disconnect them from your dashboard.
Open MCP connections