Open Source MCP Server

CloudPulse MCP

CloudPulse gives AI agents read-only cross-cloud infrastructure visibility so they can diagnose issues across AWS, Vercel, GCP, and Cloudflare without leaving the editor.

MITMCPRead-OnlyCross-Cloud
View on GitHub

Overview

CloudPulse focuses on a common AI-operations gap: agents can edit, deploy, and reason about code, but they often lack safe visibility into the cloud systems that explain runtime failures. CloudPulse exposes cloud topology, logs, service connectivity, and quota pressure through MCP tools that fit naturally into agent workflows.

The server auto-detects local credentials from AWS CLI profiles, environment variables, and platform-specific tokens. It follows a no-storage credential model and keeps diagnostic access read-only.

Why CloudPulse?

Frontend errors on Vercel require switching into AWS consoles.

Correlates Vercel and AWS CloudWatch timelines with get_correlated_logs.

AI agents cannot easily verify whether a security group blocks port 5432.

Inspects live security group rules with diagnose_service_link.

Lambda concurrency and quota pressure can stay hidden until production impact.

Warns when resources approach configured thresholds with check_resource_limits.

Topology is unclear before debugging starts.

Builds a unified active-service map in seconds with list_cloud_topology.

Supported Cloud Surface

  • AWS Lambda, RDS/Aurora, CloudWatch, Security Groups, S3, and Service Quotas
  • Vercel projects, deployments, logs, and environment-aware diagnostics
  • GCP Cloud Run, Cloud SQL, and Cloud Logging
  • Cloudflare Workers, Pages, account-scoped APIs, and Worker tail logs

Available MCP Tools

list_cloud_topology

Scans configured cloud platforms and returns a unified service map across AWS, Vercel, GCP, and Cloudflare.

get_correlated_logs

Fetches and merges Vercel and AWS CloudWatch logs into one timeline using time ranges, trace IDs, projects, and log group prefixes.

diagnose_service_link

Checks why a source service cannot reach a target resource by validating environment variables, security group rules, ports, and external reachability.

check_resource_limits

Queries quotas and flags resources nearing the configured warning threshold, defaulting to 80 percent usage.

Quick Start

Workflow

  1. Run CloudPulse with npx cloudpulse-mcp.
  2. Provide cloud credentials through local profiles or environment variables.
  3. Add the MCP server configuration to Claude Desktop, Cursor, or VS Code MCP settings.
  4. Enable the CloudPulse tools in your AI client and ask infrastructure questions naturally.

npx Configuration

Use this pattern for Claude Desktop or Cursor when running the published package directly.

{
  "mcpServers": {
    "cloudpulse": {
      "command": "npx",
      "args": ["-y", "cloudpulse-mcp"],
      "env": {
        "VERCEL_TOKEN": "<your-vercel-token>",
        "AWS_REGION": "us-east-1",
        "AWS_PROFILE": "default"
      }
    }
  }
}

AI Client Setup

Claude Desktop

Configure cloudpulse under mcpServers with npx, VERCEL_TOKEN, AWS_PROFILE, and AWS_REGION.

Cursor

Add .cursor/mcp.json to the project and run CloudPulse through npx with project-local MCP settings.

VS Code + GitHub Copilot Agent Mode

Build the project, define .vscode/mcp.json, enable CloudPulse tools, and query from Copilot Chat Agent mode.

Credentials & Security

Credential Sources

  • AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY, AWS_PROFILE, or an EC2 instance role
  • VERCEL_TOKEN and optional VERCEL_TEAM_ID
  • GOOGLE_APPLICATION_CREDENTIALS for GCP
  • CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID for Cloudflare

Security Model

  • Read-only cloud inspection by design.
  • No credential logging or credential storage.
  • Environment variables are read at call time.
  • Secrets can be referenced from shell environments instead of source-controlled config files.

VS Code MCP Configuration

For VS Code and GitHub Copilot Agent Mode, build the project first and reference the compiled server from .vscode/mcp.json.

{
  "servers": {
    "cloudpulse": {
      "type": "stdio",
      "command": "node",
      "args": ["${workspaceFolder}/dist/index.js"],
      "env": {
        "VERCEL_TOKEN": "${env:VERCEL_TOKEN}",
        "AWS_REGION": "${env:AWS_REGION}",
        "AWS_PROFILE": "${env:AWS_PROFILE}"
      }
    }
  }
}

Development

Local Setup

  1. Clone github.com/Galadriel-Tech-Solutions/cloudpulse-mcp.
  2. Run npm install.
  3. Run npm run dev to start from source with tsx.
  4. Run npm run build to compile the server to dist/.

Roadmap

  • Phase 1 completed: Vercel and AWS coverage for Lambda, RDS, CloudWatch, Security Groups, and S3.
  • Phase 2 completed: GCP Cloud Run, Cloud SQL, Logging, Cloudflare Workers, Pages, and S3 CORS checks.
  • Phase 3 planned: Pre-built diagnostic playbooks for CORS, 504 timeouts, and cold-start loops.