cctl — Universal Cloud CLI

One command for AWS, Azure, and GCP. Diagnose, fix, and manage cloud infrastructure with AI.

What is it?

cctl installs the cloudctl command — a single CLI that works across AWS, Azure, and GCP using your existing credentials. No new login. No config files to write.

Three modes, one install:

cloudctl compute list --cloud all          # query all three clouds at once
cloudctl debug "payments returning 502s"   # AI diagnoses your incident
cloudctl ask "why is prod slow?"           # AI answers cloud questions

Install

# Core CLI (AWS, Azure, GCP)
pip install cctl

# With MCP server (Claude Desktop / Cursor integration)
pip install cctl[mcp]

# With AI debug + ask commands
pip install cctl[ai]

# Everything
pip install cctl[all]

Quick Start

# First-run: auto-detects your existing AWS/Azure/GCP credentials
cloudctl init

# See all configured cloud accounts
cloudctl accounts list

# Query across all clouds at once
cloudctl compute list --cloud all
cloudctl cost summary --cloud all
cloudctl security audit --cloud all

# Target a specific account or environment
cloudctl compute list --account prod
cloudctl database list --cloud aws --env staging

Commands

Infrastructure

AI-Powered (requires pip install cctl[ai])

Setup

cloudctl debug

cloudctl debug is the flagship feature. Give it a symptom in plain English — it fetches real data from your cloud, correlates a causal timeline, runs AI analysis, detects how the affected resource was deployed, and returns IaC-aware fix steps.

cloudctl debug "payments service returning 502s since 3pm"

Diagnosing: payments service returning 502s since 3pm
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Fetching data sources...
  ✓ cloudwatch_metrics   (47 datapoints)
  ✓ cloudtrail           (23 events)
  ✓ alb_logs             (1,204 requests)
  ✓ ecs_events           (8 events)

Root Cause
  Connection pool exhausted on payments-api ECS service.
  Memory utilisation crossed 94% at 15:02 UTC — 8 minutes
  before error rate spiked. Triggered OOM kills on 3/4 tasks.

Affected Resources
  payments-api   ECS service   us-east-1
  payments-db    RDS instance  us-east-1

Deployed via: AWS CDK  ⚠ Direct changes will be overwritten on next cdk deploy

Resolution Steps
  STEP 1  Update the CDK stack — increase task memory and connection pool size:
            taskDefinition.addContainer({ memoryLimitMiB: 2048 })
            cdk deploy PaymentsStack
  STEP 2  If pipeline unavailable, scale the service directly (drift warning applies):
            aws ecs update-service --desired-count 6
  STEP 3  Fix root cause: add memory alarm at 80% to catch this earlier

[HIGH confidence — CloudWatch 24h, ALB access logs, ECS events]
Incident report saved: ~/.cloudctl/incidents/2026-04-06T15-47-payments-502.md

Deployment detection — cloudctl identifies how a resource is managed (CDK, CloudFormation, Terraform, Pulumi, Bicep, ARM, Azure DevOps, Deployment Manager, Config Connector, Cloud Build, GitHub Actions) and tailors the fix steps to your actual tooling. Detection works even without IaC tags — it inspects CloudFormation templates, ARM deployment history, GCP Deployment Manager manifests, and audit logs (CloudTrail / Activity Log / Cloud Audit Logs).

MCP Server

Works with Claude Desktop, Cursor, and any MCP-compatible client (including Bedrock-hosted Claude).

pip install cctl[mcp]
cloudctl mcp config    # prints the config block to paste into claude_desktop_config.json
cloudctl-mcp           # start the MCP server

Once connected, the AI client can query your cloud infrastructure directly:

"What's the cost breakdown for prod this month?" "List all ECS services with high CPU" "Show me the recent pipeline failures"

How It Works

• No new auth — reads your existing ~/.aws/config, ~/.azure/, and ~/.config/gcloud/
• Auto output — Rich table in terminal, clean JSON when piped (cloudctl compute list | jq)
• Multi-account — --account prod fuzzy-matches any profile name or AWS account ID
• Multi-cloud — --cloud all queries AWS + Azure + GCP in parallel
• AI is optional — all CLI commands work without AI configured; cloudctl debug and cloudctl ask require cctl[ai]

Cloud Support

Credentials

cloudctl never stores or manages credentials. It reads what you already have:

# AWS — standard profile setup
aws configure --profile prod

# Azure — existing az login session
az login

# GCP — existing gcloud session
gcloud auth application-default login

AI Provider Support

cloudctl debug and cloudctl ask work with any of these AI providers:

cloudctl config set ai.provider bedrock
cloudctl config set ai.model anthropic.claude-sonnet-4-6-v1

Links

• GitHub: https://github.com/cloudctlio/cloudctl
• Issues: https://github.com/cloudctlio/cloudctl/issues
• PyPI: https://pypi.org/project/cctl/

License

MIT