Skip to main content

Quickstart with MCP

Use the Tower MCP Server with AI coding assistants like Claude Code to build, deploy, and manage Tower apps through natural language conversations. The Tower MCP server gives your AI assistant direct access to Tower's functionality, enabling it to create apps, deploy code, and run tasks on your behalf.

Prerequisites​

  1. Tower CLI - Install with pip install tower
  2. Tower account - Sign up at app.tower.dev
  3. AI coding assistant - Claude Code, Cline, or any MCP-compatible client
  4. Active session - Run tower login to authenticate

Setup​

Run this command in terminal to add the tower mcp server

claude mcp add tower tower mcp-server

The Tower MCP server uses stdio transport, which means your AI tool automatically spawns and manages the Tower MCP server process - no manual setup or background processes required.

Clone the Tower examples repository to get started with pre-built examples and let Claude learn from them:

git clone https://github.com/tower/tower-examples

Start a Claude session in this directory:

cd tower-examples
claude

Educate Claude on how Tower works and type in the Claude session:

Read about Tower at https://docs.tower.dev/llms-full.txt

Using Tower with your AI assistant​

Once configured, you can have natural conversations with your AI assistant to build Tower apps. You can ask Claude to work with any of the examples or create new Tower apps.

Create a simple app​

Use this or similar prompt:

Create a Python app that fetches data from an API and prints a summary. 
Store all app artifacts in a new directory. Use Tower to deploy and run it. 
The app should:
- Accept a URL parameter
- Fetch JSON data from that URL
- Print a nicely formatted summary

Deploy and test​

Use this or similar prompt:

Deploy this app to Tower and run it with a test URL
Parameter syntax

If Claude has trouble passing parameters to Tower apps, remind it:

When running from the app directory, use tower run --parameter=... without the app name. 
The parameter format requires "=" after --parameter, e.g. tower run --parameter=par1=val1

Add parameters and secrets​

Use this or similar prompt:

Modify the app to accept an API key as a secret and a "location"  parameter. 
Deploy and test it.

Schedule the app​

Use this or similar prompt:

Create a schedule for this app to run every day at 8am with the location set to "London"

That's all! You now have a data retieval job that runs in the Tower cloud on a daily schedule!

What the MCP server provides​

Your AI assistant gains access to these Tower capabilities:

  • App Management - Create, deploy, and manage Tower apps
  • Run Execution - Run apps locally or remotely on Tower infrastructure
  • Parameters - Configure app parameters for different scenarios
  • Secrets - Securely manage API keys and credentials
  • Schedules - Set up automated runs with cron expressions
  • Logs - View and debug app execution logs
  • Documentation - Access Tower docs directly when needed

Tips for effective use​

  1. Be specific in prompts - Use bullet points to clearly outline what you want
  2. Start simple - Begin with basic functionality, then add complexity
  3. Ask for simplification - If code looks too complex, ask the AI to simplify it
  4. Verify outputs - Always double-check that generated code works as expected
  5. Iterate - Use the AI's ability to deploy and test to refine your app

Example workflow​

Here's a typical development flow:

  1. Describe your app - Tell the AI what you want to build
  2. Review & simplify - Check the generated code and ask for simplification if needed
  3. Deploy - Let the AI deploy the app to Tower
  4. Test - Run the app and verify the results
  5. Refine - Make adjustments based on the output
  6. Automate - Add parameters, secrets, and schedules as needed

Troubleshooting​

"Not authenticated" errors​

Make sure you've run tower login before starting your AI client with the MCP server configured.

AI not using MCP tools​

Try explicitly mentioning Tower in your prompt: "Use Tower to deploy this app" or "Use the Tower MCP server to..."

Server connection issues​

Restart your AI client after making configuration changes to ensure the MCP server is properly initialized.

Next steps​

Happy building with AI!