Honeypots play a critical role in cyber defense by emulating systems that attract and deceive attackers, allowing defenders to monitor malicious behavior in a controlled environment. However, building and maintaining realistic and diverse honeypot environments remains a labor-intensive and technically demanding task, especially when rapid deployment or customization is required. This paper proposes an automatic code generation framework that leverages Large Language Models (LLMs), named VulForge, to simplify the creation and deployment of services within honeypot environments. The framework is capable of generating deployment scripts that install vulnerable applications, simulate known software vulnerabilities, reducing manual effort while supporting scalable and realistic honeypot environments. Across three experimental scenarios, VulForge achieved a 100% deployment success rate (DSR) on common web vulnerabilities, demonstrated full manual and tool-based interaction success rates on service-specific vulnerabilities, and successfully automated the deployment of real-world CVEs with a 50% DSR despite their inherent complexity. These results highlight VulForge's effectiveness in minimizing deployment overhead while ensuring realism and demonstrate its potential to advance scalable and adaptive automated honeypot systems.