MCP Quick Start

Get Station agents running in Claude Desktop (or Claude Code) in under 5 minutes. This is the fastest way to start using Station’s deployable sub-agents.

The 5-Minute Setup

1. Install Station Binary

# Quick install (recommended)
curl -fsSL https://raw.githubusercontent.com/cloudshipai/station/main/install.sh | bash

# Verify installation
stn --version

2. Initialize Station

stn init

This creates your Station workspace and initializes the local database.

3. Create and Load an MCP Configuration

Create a filesystem MCP config directly in your Station environment:

Copy this config to ~/.config/station/environments/default/template.json:

{
  "description": "Essential filesystem operations with MCP server integration",
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y", 
        "@modelcontextprotocol/server-filesystem@latest",
        "{{ .ALLOWED_PATH }}"
      ]
    }
  },
  "name": "filesystem-updated"
}

Create variables file:

# Create variables.yml with the template variable
echo "ALLOWED_PATH: /home/user/projects" > ~/.config/station/environments/default/variables.yml

Sync to discover tools:

stn sync

Output:

╭─────────────────────────────╮
│                             │
│  🔄 MCP Configuration Sync  │
│                             │
╰─────────────────────────────╯

🔍 Scanning file-based configs in environment 'default'...
Processing MCP template: filesystem-updated
   🖥️  Processing server: filesystem
      Command: npx [-y @modelcontextprotocol/server-filesystem@latest /home/user/projects]
   ✅ Created MCP server: filesystem

✅ MCP SUCCESS: Discovered 14 tools from server 'filesystem'
   🔧 Tool 1: __read_text_file
   🔧 Tool 2: __list_directory  
   🔧 Tool 3: __directory_tree
   🔧 Tool 4: __write_file
   🔧 Tool 5: __edit_file
   [... 9 more tools discovered]

✅ Sync completed successfully!

4. Add Station to Claude Desktop

Add this to your Claude Desktop MCP configuration file:

Location of config file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this configuration:

{
  "mcpServers": {
    "station": {
      "command": "stn",
      "args": ["stdio"]
    }
  }
}

5. Restart Claude Desktop

Close and reopen Claude Desktop. You should now see Station tools available.

Start Using Station Agents

Once connected, you can immediately start using Station agents through Claude Desktop:

Create Your First Agent

You: "Create a code review agent that can analyze files and provide feedback"

Claude Response:

{
  "agent": {
    "description": "A basic agent that reviews code files and provides feedback",
    "environment_id": 1,
    "id": 53,
    "name": "Simple Code Reviewer"
  },
  "message": "Agent 'Simple Code Reviewer' created successfully",
  "tool_assignment": {
    "assigned_count": 2,
    "assigned_tools": ["__read_text_file", "__list_directory"],
    "status": "success"
  }
}

Claude automatically selected the right tools from your MCP servers!

Run Your Agent

You: "Run my code review agent to analyze the current project structure"

Agent Response:

I inspected the repository root and analyzed the project structure...

What I see:
- Project identity: This is a Go-based runtime/CLI project called 'Station'  
- Language/build: go.mod, go.sum — Go project with cmd/, internal/, pkg/ layout
- Runtime assets: stn — built CLI binary, station.db — SQLite database
- Deployment: docker/, Dockerfile, .goreleaser.yml — containerized deployment

Brief summary: This is a Go-based, self-hosted runtime that provides a 
lightweight platform for deploying and running 'sub-agents'...

Full comprehensive analysis with complete tool usage - nothing truncated.

List Your Agents

You: "List all my Station agents"

Response: Shows all your created agents with IDs, names, descriptions, and tool assignments.

What Just Happened?

In these 5 steps, you’ve:

  1. Installed a lightweight Station runtime and initialized your workspace
  2. Loaded MCP tools (filesystem operations) that agents can use
  3. Connected Station to Claude Desktop via MCP protocol
  4. Created your first deployable sub-agent through natural conversation
  5. Executed the agent and saw complete, untruncated responses

Your agents can now combine tools from multiple MCP servers and be deployed across environments as .prompt files!

Next Steps

Troubleshooting

Station not showing up in Claude Desktop?

  1. Verify stn stdio works: echo '{"jsonrpc":"2.0","method":"tools/list","id":1}' | stn stdio
  2. Check Claude Desktop config file syntax with a JSON validator
  3. Make sure Claude Desktop was fully restarted after config changes

No MCP tools available for agent creation?

  1. Copy a template config to your environment directory: ~/.config/station/environments/default/template.json
  2. Add any required variables to variables.yml
  3. Sync the configuration: stn sync
  4. Restart Claude Desktop after syncing new tools

Agent creation failing?

  1. Make sure you have MCP tools loaded (agents need tools to be useful)
  2. Verify Station initialization completed: stn status
  3. Check if the environment exists: stn mcp environments list

You’re now ready to deploy intelligent sub-agents through Claude Desktop!

The agents you create can handle sensitive internal tasks like database monitoring, deployment automation, and security scanning - all while maintaining secure access controls and audit trails.