Using with AI Agents

jack is the deployment platform for Claude Code. Here's how to get the most out of it.

Claude Code

Claude Code is the primary agent for jack. Every project includes context files (AGENTS.md, CLAUDE.md) and there's a built-in MCP server with 27 tools for direct deployment and service management.

Quick Start

Install jack and configure MCP in one command:

curl -fsSL docs.getjack.org/install.sh | bash

Start a new Claude Code session (or /mcp reconnect jack), then ask Claude to deploy something.

No terminal? Use the remote MCP server from Claude Desktop or Claude.ai:

https://mcp.getjack.org/mcp

The remote server has 10 tools (deploy, projects, databases, logs, files). Install locally for the full 27-tool experience.

Authentication

Agents need to authenticate before deploying. There are three ways, depending on your setup.

Pre-existing token (simplest)

If you already have an API token, set it as an environment variable. No login needed.

export JACK_API_TOKEN=jkt_abc123...

Create tokens with jack tokens create.

Magic auth (headless, no browser)

For agents that can send/receive email, or for human-supervised setups where a human can read the code from email.

Two-step flow (fully non-interactive):

jack login --email [email protected]

This sends a 6-digit verification code to the email and exits. Then:

jack login --email [email protected] --code 847291

This verifies the code, registers the account, and creates a jkt_* API token. The CLI is authenticated for all future commands. The token is also printed to stdout for programmatic capture.

Interactive flow (human-supervised):

If running in a terminal, jack login --email [email protected] will prompt for the code inline — the human reads it from email and pastes it.

Environment variables:

Variable	Purpose
`JACK_EMAIL`	Same as `--email` flag — triggers magic auth automatically
`JACK_API_TOKEN`	Skip login entirely, use an existing token
`CI=true`	Forces non-interactive mode (no prompts)

Browser login (interactive humans)

Running jack login without --email opens a browser for OAuth login. This is the default for interactive terminal sessions. If no browser is available, it fails fast with instructions to use --email instead.

MCP Server

jack includes a built-in Model Context Protocol server that lets AI agents deploy and manage projects directly.

Setup

The install script configures MCP automatically for all detected editors:

Claude Code (~/.claude.json)
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json)
Cursor (~/.cursor/mcp.json)

To manually configure or start the MCP server:

jack mcp install   # Write config to detected editors
jack mcp serve     # Start MCP server manually

Available Tools

Project Management

Tool	Description
`create_project`	Create and deploy a new project from a template or fork
`deploy_project`	Deploy the current project
`get_project_status`	Check deployment status, URL, and last deploy time
`list_projects`	List all projects (filter by local, deployed, or cloud)

Forking and Templates

The create_project tool supports three template modes:

Built-in:    template: "api"                 → starts from a builtin template
Fork own:    template: "my-project"          → forks your own deployed project
Fork public: template: "alice/cool-api"      → forks a published project

All modes create a fully deployed, independent copy.

Database

Tool	Description
`create_database`	Create a database for a project
`list_databases`	List project databases
`execute_sql`	Execute SQL queries (read-only by default, `allow_write` for mutations)

Vector Search

Tool	Description
`create_vectorize_index`	Create a vector index for similarity search
`list_vectorize_indexes`	List project indexes
`get_vectorize_info`	Get index metadata (dimensions, metric, count)
`delete_vectorize_index`	Delete an index

Storage

Tool	Description
`create_storage_bucket`	Create a storage bucket
`list_storage_buckets`	List project buckets
`get_storage_info`	Get bucket info
`delete_storage_bucket`	Delete a bucket

Observability

Tool	Description
`start_log_session`	Start a 1-hour real-time log session
`tail_logs`	Collect a short sample of live log events

Scheduling (Cron)

Tool	Description
`create_cron`	Create a cron schedule (minimum 15 min interval)
`list_crons`	List project cron schedules
`delete_cron`	Delete a cron schedule
`test_cron`	Validate expression and show next 5 scheduled times

Custom Domains

Tool	Description
`list_domains`	List all custom domains and their status
`connect_domain`	Reserve a custom domain slot
`assign_domain`	Assign a domain to a project
`unassign_domain`	Unassign a domain from its project
`disconnect_domain`	Fully remove a domain

Multi-Project Management

Most tools accept a project_path parameter, so an agent can manage multiple projects in a single session. Tool calls can run in parallel — a single prompt can launch several projects simultaneously, each returning a live URL.

From there, you can keep working across all of them: add databases, seed data, configure cron jobs, check logs — all without switching context.

create_project    name: "users-api", template: "api"
create_project    name: "admin-dash", template: "miniapp"
deploy_project    project_path: "/projects/users-api"
execute_sql       project_path: "/projects/users-api", sql: "CREATE TABLE users ..."
create_cron       project_path: "/projects/admin-dash", expression: "0 * * * *"
tail_logs         project_path: "/projects/users-api"

Other Agents

jack also works with other AI coding agents (Codex, Cursor, etc.) via CLI commands and --json output. The MCP server is compatible with any agent that supports the Model Context Protocol.

AGENTS.md

Every jack project includes an AGENTS.md file with instructions for AI agents. This helps your coding agent understand your project structure.

Preferred Agent

jack can automatically open your preferred coding agent after creating a project:

Set your preferred agent:

jack agents prefer claude-code

Available: claude-code, codex.

Now jack new will open your agent automatically:

jack new my-app

Tips for AI-Assisted Development

Let the agent deploy — agents can run jack ship directly
Use MCP tools — agents with MCP access can create, deploy, query, and manage without the CLI
Fork as a starting point — use create_project with a template to fork existing projects
Go wide — launch multiple projects in one prompt, manage them all from one session
Check AGENTS.md — customize it for your project's specific patterns