A team of developers has built a custom Model Context Protocol (MCP) server to integrate Jira, Confluence, and GitLab within their corporate environment. The project was completed in two weeks and currently supports nearly 100 tools, addressing security and scalability limitations found in existing open-source solutions.

The architecture uses a centralized Java-based backend deployed in Kubernetes, utilizing Spring Boot 4.1, Spring AI 2.0, and PostgreSQL 17. It manages user authentication via Keycloak integrated with Active Directory, while sensitive tokens are encrypted and stored using Vault. The system supports both streamable-http and stdio transport protocols.

The design prioritizes security by isolating user tokens and implementing granular access control. Each user manages their own credentials through a web interface, and the system ensures that raw encryption keys never leave the Vault. The team also developed a native stdio-proxy to support clients lacking HTTP authentication capabilities.

The development process highlighted significant challenges with API specifications and null-value handling in generated clients. To ensure reliability, the team implemented a mandatory testing phase against real system instances, as unit tests using incomplete specifications failed to catch critical integration errors.