Presenters

Source

Revolutionizing AI Integration: Toyota Connected’s Journey with Model Context Protocol 🚀

In the fast-paced world of tech, staying ahead means constantly innovating and finding smarter ways to work. Toyota Connected, a powerhouse behind the massive 12.5 million vehicles on their Drive Link platform and processing a staggering 29 billion transactions monthly, has embarked on an exciting journey to supercharge their generative AI adoption. Their secret weapon? The Model Context Protocol (MCP) and a brilliant strategy to make its integration seamless for their developers.

Let’s dive into how they’re tackling AI integration, turning complex challenges into opportunities for enhanced productivity and innovation! ✨

The AI Bottleneck: The Pain of Manual Context Gathering 😩

Before the MCP revolution, Toyota Connected’s developers faced a significant hurdle. Feeding AI models the right information was a manual, time-consuming, and error-prone process. Imagine:

  • Juggling Disparate Sources: Developers had to painstakingly gather crucial context from a multitude of places – Jira for tasks, GitLab/GitHub for code, and Confluence for documentation.
  • The Copy-Paste Peril: This manual collection often involved copy-pasting, a breeding ground for mistakes and outdated information.
  • Hallucinations Galore: Outdated or incomplete context directly led to AI “hallucinations,” where the AI generated incorrect or nonsensical outputs, hindering progress.

This manual grind was simply not sustainable for a company operating at Toyota Connected’s scale.

The MCP Solution: Standardizing AI Context with Ease 🛠️

Enter the Model Context Protocol (MCP). This open standard is a game-changer, allowing AI applications to consistently and modularly gather context and interact with external applications. To make this even more accessible, Toyota Connected built a sustainable, self-service template for deploying MCPs. This wasn’t just about adopting a new technology; it was about creating an ecosystem that prioritized:

  • Ease of Use: Lowering the barrier to entry for application developers.
  • Scalability: Handling massive growth and complexity.
  • Security: Ensuring data is protected.
  • Sustainability: Building for the long haul.

The Magic Behind the MCP Deployment Process ✨

The team has streamlined the deployment of MCPs into a remarkably user-friendly experience:

  • The Four-Field Entry Point: Developers start by simply providing four essential pieces of information: owner, department, name, and description. That’s it!
  • Automated Infrastructure Powerhouse: The magic happens behind the scenes. The process automatically spins up GitLab projects, configures CI/CD pipelines, builds containers, and leverages Customize for seamless Argo CD file management.
  • Effortless Deployment: The automated workflow pushes configurations to Argo CD, which then handles the deployment of the MCP application. This even includes integrating external DNS and the External Secrets Operator for secure credential management.

The underlying technology stack is a robust mix of industry-leading tools: OpenTofu for infrastructure as code, Kubernetes for container orchestration, Argo CD for continuous delivery, Prometheus and OpenCost for cost visibility, and Datadog for comprehensive monitoring.

Backstage Integration: Bringing MCPs into the Developer’s Workflow 👨‍💻

The true power lies in how these MCPs are surfaced to developers. By integrating MCPs into Backstage, Toyota Connected provides a central hub for all things AI context:

  • Dynamic Catalog Entries: As soon as an MCP is deployed, it’s automatically registered in the Backstage catalog with a kind: mcp type, making it instantly discoverable.
  • The Dedicated “MCP Entity Tab” 💡: This is where the magic really shines. A custom tab within Backstage displays critical information about each MCP server, acting as a de facto “API spec” for MCPs. Developers can see:
    • The tools the MCP interacts with.
    • The prompts it’s designed to handle.
    • The resources it provides.
  • Conquering Authentication Challenges 🛡️: A significant hurdle was securing MCP servers with OAuth or token authentication, which initially prevented Backstage from directly querying them.
  • Dynamic Documentation Generation FTW! 🚀
    • Initial Approach (FastMCP Inspect): They began by using FastMCP Inspect to generate local JSON metadata, which Backstage could then consume.
    • The Real-Time Solution: To overcome the limitations of static JSON, they developed a dedicated endpoint within their infrastructure automation. This endpoint dynamically runs the MCP inspection logic, providing real-time data to Backstage.
    • Monorepo Fallback: For teams using monorepos outside their automated workflow, Backstage gracefully falls back to consuming JSON metadata. The onus is then on the developers to ensure this metadata is kept up-to-date.
    • Guidance for the Uninitiated: If neither a docs endpoint nor JSON metadata is available, Backstage provides clear guidance on how to generate the necessary JSON, ensuring no developer is left behind.

The MCP Hub Plugin: Centralizing Discovery and Onboarding 🌐

To further accelerate MCP adoption and make finding the right AI context even easier, Toyota Connected developed the “MCP Hub” plugin within Backstage:

  • Internal Registry: Your Curated AI List: This section highlights curated MCP servers within the organization. It’s a perfect starting point for users who might feel overwhelmed by the broader catalog.
  • Community Picks: The Wisdom of the Crowd 🧑‍🤝‍🧑: This tab showcases community-contributed MCPs, featuring exciting examples like Context 7, Playwright, and AWS documentation MCPs.
    • Information Security Integration: This tab even integrates with Information Security, allowing them to rate community MCPs. This consolidates security reviews and user recommendations, building trust and confidence.
  • Sonnet for Metadata Magic 🪄: For community MCPs, they leverage Sonnet 4.5 to automatically generate metadata (title, summary, tags) directly from a GitHub link. This dramatically reduces manual form-filling for users.
  • Kickstarter Guide: Demystifying AI Adoption 🚀: This section provides step-by-step guides for setting up MCPs with popular tools like Cloud Code and GitHub Copilot. The goal is clear: reduce AI hesitancy and enthusiastically encourage adoption.

Key Learnings and the “Why”: Building for the Future, Today 🎯

The overarching theme from Toyota Connected’s journey is proactive platform engineering. They built the infrastructure for generative AI before the demand became overwhelming. Their philosophy is simple yet powerful:

  • Adoption is King: Platform engineering success isn’t about forcing users; it’s about providing tangible value that developers naturally gravitate towards.
  • Automate the Pain Points: Continuously identifying and automating tedious or error-prone tasks frees up developers to focus on what truly matters – innovation.
  • Roadmaps Through Calculated Risks: They are not afraid to take calculated risks on emerging technologies like MCPs, understanding that this is how future-proofing happens.

The results speak for themselves: consistent quarter-over-quarter growth in adoption rates.

Future Vision and Q&A Insights: What’s Next? 🔮

The conversation didn’t stop there! The team shared exciting glimpses into their future vision:

  • Deeper Backstage Integration: Imagine MCPs being even more deeply embedded within Backstage, with Backstage itself leveraging MCPs for enhanced contextual awareness.
  • UI Simplicity is Key: They acknowledge that Backstage can sometimes feel cluttered. The focus is on dynamically showing relevant tabs – for instance, an MCP tab only appearing for MCP entities – to create a cleaner, more intuitive user experience.
  • Fortifying Security: Security remains paramount. Ongoing work includes robust OAuth implementation for MCPs and automating secret management to prevent “MCP poisoning.” The emphasis is on internal network security and establishing standardized best practices in collaboration with infosec teams.
  • Abstraction of Point Products: The ultimate long-term vision is for MCPs to abstract away underlying tools from end-users, leading to an even more seamless and productive developer experience.

Toyota Connected’s approach to generative AI integration is a masterclass in platform engineering, demonstrating how thoughtful design, automation, and a deep understanding of developer needs can unlock incredible potential. Their MCP journey is a compelling story of innovation, efficiency, and a clear vision for the future of AI-powered development.

Appendix