Station: Open Source MCP Runtime for AI Agents | Self-Hosted DevOps Automation

Remember when we tried to sell DevOps agents to enterprises? Yeah, that didn't go as planned. Turns out, nobody wants to send their AWS credentials to some random startup's cloud. Who would've thought? So we did what any rational team would do - we said screw it and open-sourced the whole thing. Meet Station: an MCP runtime that lets you build AI agents that actually run in YOUR infrastructure. 137 stars in week one. Not bad for a pivot.

The Reality Check That Changed Everything

Picture this: We're in a Zoom call with a Fortune 500 CISO. We're demoing our amazing cloud-hosted DevOps agents. Everything's going great until he asks, 'So our database passwords go through your servers?' The silence was deafening. That's when it hit us - we were solving the wrong problem. Teams don't need another SaaS tool. They need agents they can trust, running where their data lives.

Station vs Other AI Agent Platforms

Look, I'll save you the marketing BS. Here's the actual difference:

Feature	Station	Cloud Agents	Traditional CI/CD
Credentials	Never leave infrastructure	Sent to 3rd party	Local only
MCP Support	Native, 30+ tools	Limited	None
Install Time	30 seconds	Hours	Days
Self-Hosted	✅ Yes	❌ No	✅ Yes
AI Integration	Claude, Cursor, Any LLM	Vendor lock-in	None
Cost	Free, Open Source	$$$$/month	$$$/month

What is a Deployable Sub-Agent?

Here's the thing that blew my mind when our engineer showed me: a Station agent is literally just a text file. Not a Docker container. Not a complex YAML manifest. Just a .prompt file that tells the AI what to do, plus whatever MCP tools it needs. That's it. You can version it, email it, hell, you can tweet it if it's short enough.

---
model: "gemini-2.5-flash"
config:
  temperature: 0.3
input:
  schema:
    userInput: string
    project_type: string
metadata:
  name: "planner"
  description: "A planning agent for DevOps tasks"
---

{{role "system"}}
You are an expert planning agent specialized in DevOps automation...

{{role "user"}}
{{userInput}}

See that? That's a production-ready agent. Copy it, paste it, run it. We've had teams deploy their first agent faster than it takes to read the AWS documentation for Lambda.

Quick Start: Install Station in 30 Seconds

# Install Station
curl -fsSL https://raw.githubusercontent.com/cloudshipai/station/main/install.sh | bash

# Initialize
stn init

# Sync configuration
stn sync

# Connect to Claude Desktop
echo '{
  "mcpServers": {
    "station": {
      "command": "stn",
      "args": ["stdio"]
    }
  }
}' >> ~/.claude_desktop_config.json

30+ DevOps Tools Already Converted to MCPs

Remember when I said we open-sourced everything? We meant EVERYTHING. We spent months converting 30+ DevOps tools into MCP servers. Trivy, Terraform, Kubectl - all the hits. One binary. Ship CLI installs them all. You're welcome.

Security Tools: Trivy (22.8k⭐), Snyk (4.9k⭐), Checkov (7.1k⭐) - vulnerability scanning and compliance
Infrastructure: Terraform (42.5k⭐), Kubectl (28.1k⭐), Helm (26.8k⭐) - infrastructure management
CI/CD: GitHub Actions, Jenkins, ArgoCD - pipeline automation and monitoring
Cloud Platforms: AWS, GCP, Azure - cost optimization and resource management
Monitoring: Prometheus, Grafana, Datadog - observability and alerting

Real Production Use Cases

Okay, real talk - here's what people are actually building with this thing (and yes, these are real production deployments, not 'we tested it once' examples):

Security & Compliance Automation

# security-scanner.prompt
---
model: "claude-3-sonnet"
metadata:
  name: "security-scanner"
  description: "Automated CVE scanning with auto-remediation"
---
Scan all containers for vulnerabilities.
Generate fix PRs for critical issues.
Notify team via Slack for manual review items.

What happened: One team went from 4-hour incident response to 30 minutes. Their security guy literally texted me 'I actually got to eat lunch today.' That's the metric that matters.

Cost Optimization Agent

# cost-optimizer.prompt
---
model: "gpt-4"
metadata:
  name: "cost-optimizer"
  description: "Identify and eliminate cloud waste"
---
Analyze AWS Cost Explorer data.
Identify idle EC2 instances, unattached EBS volumes.
Generate Terraform code to rightsize resources.
Calculate projected savings.

The damage: $180k saved in 3 months. The CFO asked if the AWS bill was wrong. Nope, just an agent that actually understands what 'idle resources' means and isn't afraid to terminate them.

Advanced Features for Enterprise Teams

Interactive Development Playground: Browser-based agent testing with `genkit start -- stn develop`
Environment Management: Seamless dev/staging/production deployments with isolated configurations
Bundle Deployment: Package agents with dependencies for easy distribution
Web Management UI: Monitor execution, manage agents, test tools at http://localhost:8585
Secure Variable Management: Declarative variables with vault integration, never exposed to AI

MCP Server Integration

The best part? Station works with ANY MCP server. Got a weird internal tool? Make it an MCP. That janky bash script Bob wrote in 2019? MCP it. Here's how dead simple the config is:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem@latest", "{{ .ROOT_PATH }}"]
    },
    "aws": {
      "command": "npx",
      "args": ["-y", "@aws/mcp-server", "--region", "{{ .AWS_REGION }}"]
    }
  }
}

Why Open Source Matters for AI Agents

Here's why open source actually matters (beyond the warm fuzzy feeling). When your agent is handling production database credentials at 3am, you need to know EXACTLY what it's doing. No black boxes. No 'trust us.' Just code you can read, audit, and modify:

See everything: Every line of code. Every API call. Every decision tree. Your compliance team will love you.
Own everything: Fork it. Break it. Fix it. Make it yours. We literally can't stop you.
Pay nothing: No per-agent fees. No API charges. No 'contact sales for pricing.' It's free. Forever.
Deploy anywhere: Your laptop. Your data center. That Raspberry Pi in your closet. If it runs Linux, it runs Station.
Trust nobody: Not even us. The code is there. Audit it. We dare you to find something sketchy.

Join the Station Community

The coolest part about open sourcing? The community showed up. Like, actually showed up. People are building insane stuff and sharing it. Join the chaos:

⭐ Star us on GitHub - 137+ stars and growing
📚 Read the Documentation - Complete setup and API guides
🎯 Browse Agent Templates - 23+ ready-to-use templates
🔧 View MCP Tools - 30+ DevOps tools as MCPs
💬 Join Discord - Get help, share agents, contribute

What's Next for Station

We're not done. Not even close. Here's what's coming (and no, these aren't 'maybe someday' features - we're literally coding this stuff right now):

Next month: Kubernetes operator because apparently everyone runs k8s now
Also next month: Vault integration because storing secrets in YAML is apparently 'bad practice'
Q2: Team features so you can blame someone else when the agent goes rogue
Q2: One-click cloud deploys for people who think curl | bash is too many steps
Later: Agent marketplace where you can sell that hacky script you wrote at 2am

Get Started Today

Look, I could give you the whole sales pitch about how Station will revolutionize your DevOps workflow and synergize your cloud native blockchain AI whatever. But here's the truth: it's just a really good tool for running AI agents without sending your secrets to the cloud. That's it. That's the pitch.

# Install Station now
curl -fsSL https://raw.githubusercontent.com/cloudshipai/station/main/install.sh | bash

# Or view the source first
git clone https://github.com/cloudshipai/station
cd station
make install

We tried to build a SaaS. The market said no. So we built something better - something you actually own. Station is live, it's free, and it's yours. Your credentials never leave your servers. Your agents run where you want them. And if you don't like something? Change it. We literally cannot stop you. That's not a bug, that's the whole point. Welcome to Station. Now go build something cool and tell us about it.