Your teams can now ask EventCatalog about your architecture, events, services, and schemas directly in Slack.
You know the pattern. Someone on your team needs to know which services consume the OrderCreated event. They check the wiki. Not there. They search Confluence. Wrong version. They ping three different people in Slack asking if anyone knows.
I'm excited to share that we've improved and simplified our MCP server for EventCatalog.
Previously, to use EventCatalogs MCP server you had to install and run the standalone @eventcatalog/mcp-server package separately, the previous MCP server would use the llms.txt file to provide context to the LLM. Although this worked, it was not the most efficient way to use EventCatalog with MCP clients...
We have reviewed the MCP server and made it much more efficient and easier to use.
Every EventCatalog instance now includes a built-in MCP server at /docs/mcp/ and now uses more performant ways to provide your models with the information they need.
This means you can connect your preferred MCP clients directly to your EventCatalog instance and start asking questions about your architecture in natural language.
EventCatalog is documentation for both humans and AI. Your teams browse the catalog to understand services, events, and domains. Your AI tools can now do the same.
The MCP server gives LLMs and MCP clients (Claude Desktop, Cursor, Windsurf, VS Code) direct access to your architecture documentation. Ask questions about message schemas, trace service dependencies, or analyze change impact.
The AI works with the same source of truth your team uses.
This is the best of both worlds. Humans get a visual catalog they can browse and search. AI gets structured access to query, filter, and reason about your architecture. Both stay in sync because they read from the same place.
To get started, you need to update your EventCatalog instance to the latest version and configure it to run in SSR mode.
First you need to update your EventCatalog instance to the latest version.
npm install @eventcatalog/eventcatalog@latest
Then you need to update your eventcatalog.config.js file to use SSR mode. (mcp requires SSR mode)
export default {
output: 'server',
// ... other config options
}
Once EventCatalog is running you can connect your preferred MCP client to your EventCatalog instance at /docs/mcp/.
To get started, see our MCP server documentation.
Once connected, you can ask questions about your architecture in natural language. This can be useful for:
Here are some examples:
"What's the blast radius if I change the OrderCreated event?"
The OrderCreated event (v2.0.0) is consumed by 4 services:
Service Team Impact InventoryService Platform Updates stock levels NotificationService Communications Sends order confirmation AnalyticsService Data Tracks order metrics BillingService Finance Initiates payment flow Total blast radius: 4 services across 4 teams. Consider coordinating with Platform and Finance teams before making breaking changes.
"Which services does the Payment team own?"
The Payment team owns 3 services:
- PaymentGateway (v1.2.0) - Processes credit card transactions
- RefundService (v1.0.0) - Handles refund requests and chargebacks
- PaymentWebhooks (v2.1.0) - Receives payment provider callbacks
These services produce 5 events and consume 3 events total.
"Show me the schema for UserRegistered"
Here's the UserRegistered event schema (v1.0.0):
{
"userId": "string (UUID)",
"email": "string (email format)",
"registeredAt": "string (ISO 8601 datetime)",
"source": "enum: web | mobile | api",
"marketingConsent": "boolean"
}This event is produced by AuthService and consumed by NotificationService, CRMSync, and AnalyticsService.
"Explain the checkout business flow"
The Checkout flow consists of 5 steps:
- Cart Validation - CartService validates items and availability
- Price Calculation - PricingService applies discounts and taxes
- Payment Processing - PaymentGateway charges the customer
- Order Creation - OrderService creates the order record
- Fulfillment Trigger - Emits OrderCreated to start shipping
Average completion time: 2.3 seconds.
The MCP server includes 15 tools for querying your architecture and 9 resources for bulk data access.
View all tools and resources →
The standalone @eventcatalog/mcp-server package still works, but we plan to deprecate it in a future release. The built-in server is faster, uses less context, and requires no external setup. We recommend migrating to the built-in MCP server when possible.
The built-in MCP server simplifies deployment by eliminating external server management. It provides faster queries through direct content collection access and supports custom tools for domain-specific integrations.
The built-in MCP server requires SSR mode and a Scale license. Get a free trial at eventcatalog.cloud.
Learn more:
Get help:
EventCatalog already gives you auto-generated visualizations: Entity Maps, Interaction Maps, and domain views that update as your catalog changes. These are great for understanding relationships between your resources.
But sometimes you need your own diagrams. Target architecture plans. Event storming results. Sequence flows from Miro. C4 diagrams from IcePanel. The stuff that lives in scattered boards, Confluence pages, or that folder someone created two years ago.
EventCatalog 3.3.0 lets you bring those diagrams to your documentation.
Your custom diagrams are now first-class, versioned resources in your catalog. Bring them in from any tool, version them, link them to your domains and services, and even ask AI about them.
I'm excited to announce Custom Tools for EventCatalog Assistant, allowing you to extend the AI with your own integrations and bring real-time data directly into your architecture conversations.
Your documentation is valuable, but it only tells part of the story. With custom tools, you can now ask EventCatalog questions like "Is OrderService healthy?" or "Who's on-call for PaymentService?" and get answers based on live data from your production systems.
We’re excited to announce that EventCatalog now supports seamless integration with OpenAI.
This new feature allows you to harness the power of OpenAI models and embeddings directly within your catalog.
With this new feature you can bring your own OpenAI keys to EventCatalog and get architecture insights in seconds using EventCatalog Chat.
EventCatalog Chat transforms how your team interacts with your event-driven architecture. Instead of digging through documentation or tracking down who produces what, you can simply ask. Want to know what events exist, how schemas are structured, or who’s publishing and consuming them? Just chat with your catalog and get instant answers—saving you time and effort.
EventCatalog Chat can also generate code on the fly and supports custom prompts, allowing you to define reusable queries aligned with your organization's best practices and governance standards. Whether it's architecture insights or scaffolding code, your team can move faster with AI-powered support built right into your catalog.
Previously, EventCatalog Chat relied solely on browser-based models—open-source models that ran entirely within the browser environment. While this offered a lightweight and privacy-conscious option, we heard your feedback loud and clear. Now, we’re excited to introduce support for bringing your own OpenAI models to EventCatalog, unlocking even more powerful capabilities for exploring and understanding your architecture.
To get started you can read the EventCatalog Chat Guide or watch the video below.
We're introducing a powerful new concept in EventCatalog Chat: Bring Your Own Prompts.
This feature allows teams and organizations to define custom prompts tailored to their own standards, best practices, and workflows. With predefined prompts, you can guide how EventCatalog Chat responds—ensuring consistency, compliance, and faster results across your team.
Here are a few examples of what you can do:
With BYO Prompts, your event-driven architecture just got a whole lot smarter—and more tailored to you.
You can also make your prompts dynamic, allowing EventCatalog Chat to ask users for input before sending the prompt to your OpenAI model.
For example, if you want to generate a JSON schema that follows your company’s best practices, you can first prompt the user to enter the event name they’re working with. This adds flexibility and interactivity to your custom prompts—making them more useful across different teams and scenarios.
When the user submits the form, the AI model uses their input to generate a JSON schema—fully aligned with the standards and conventions your company defines. This ensures consistency while saving your team valuable time on repetitive tasks.
Custom prompts can also capture multiple inputs, making them even more powerful.
For instance, you can prompt users to select an event from a generated list based on your catalog, and choose a programming language from a predefined list. This gives your team a flexible and guided way to generate highly specific outputs—like language-specific producer or consumer code for a particular event.
The prompt is then sent to your OpenAI model, which generates the code based on the inputs provided. In the example above, a TypeScript Kafka producer is generated using the Delivery Failed event as the foundation—automatically tailored to your architecture and development standards.
Check out our EventCatalog Chat Guide to get started. If you have any questions or want to connect with the community, feel free to join us on Discord.
We’re just getting started. Up next, we’re working on expanding model support in EventCatalog—including the ability to configure additional providers like Anthropic, giving you more flexibility and choice in how you power your architecture insights.
We’re also building features that let you ask questions directly within your documentation pages, so teams can get instant, contextual answers while exploring or working on your event-driven systems—saving even more time and reducing friction.
Have questions or feedback? Join us on Discord—we’d love to hear from you.
Want a tailored walkthrough? Book a custom demo—we’re happy to help!
Yes, you can configure which OpenAI model is used by updating the model property in the eventcatalog.config.js file.
You can also configure which embedding model is used.
Yes, it is secure. If you want to use your own OpenAI keys you can build EventCatalog as a server and bring your own OpenAI keys.
A new server output type is available in EventCatalog. This will output a server application that you can run locally or in a cloud provider.
You can read more about hosting a server in the EventCatalog documentation.
