MCP registries in Catalog (tech preview)

Tech Preview and uses: Catalog
This feature is currently in Tech Preview and should not be used in a production environment.
Related Documentation
Incompatible with
on-prem

What is an MCP Registry?

An MCP Registry is a centralized publishing and discovery endpoint for MCP servers within your organization.

As organizations experiment with AI agents, MCP servers are often embedded directly into agent code, stored in local configuration files, or scattered across repositories. Over time, this can lead to MCP sprawl, making it difficult to understand:

  • What MCP servers exist
  • Which agents should use which servers
  • How those servers should be governed

MCP Registries extend Konnect Catalog to provide a structured, standards-based way to register and manage MCP servers, helping Platform Teams maintain visibility and control as AI adoption scales.

This feature is built on top of the MCP Registry API specification defined by Anthropic’s open source project. The specification allows flexibility in how registries are structured, so you can model them according to your organization’s governance needs.

Enable MCP Registries in Catalog

MCP Registries in Catalog are currently available in tech preview via Konnect Labs.

  1. In Konnect, click your organization dropdown and select “Manage organizations”.
  2. Click the Labs tab.
  3. Click Catalog - MCP Registry.
  4. Click Enable feature.

You can access MCP Registries by doing the following:

  1. In the Konnect sidebar, click Catalog.
  2. Click the MCP registries tab.

    If you’re using Catalog Classic, click MCP Registries in the sidebar instead.

Create an MCP Registry

Add an MCP server

An MCP server represents an agent-facing service definition. It describes:

  • What the server does
  • Its version
  • How agents can connect to it

Packages and remotes

MCP servers are defined independently from a single deployment. They may include multiple delivery paths.

Packages

Packages describe run-it-yourself distribution options, such as:

  • npm packages
  • PyPI packages
  • OCI artifacts

These allow teams to install and run the MCP server within their own agent environments.

Remotes

Remotes describe hosted MCP server endpoints that agents can connect to over the network, such as:

  • streamable-http
  • sse

An MCP server can define multiple packages and multiple remotes simultaneously. This allows organizations to support different runtime environments without duplicating server definitions.

Access and authentication

Registry endpoints assume a Konnect authentication context. Only authenticated clients with appropriate access tokens can query the registry URL:

curl -X GET "https://klabs.us.api.konghq.com/v0/mcp-registries/internal-mcp-registry/v0.1/servers" \
  -H "Authorization: Bearer $KONNECT_TOKEN"

What’s next

We are continuing to evolve MCP Registries alongside the broader MCP ecosystem.

Planned enhancements include:

  • Linking Catalog MCP servers to MCP servers created in AI Gateway 2.0
  • Additional governance and lifecycle controls

Because the MCP specification is still evolving, we are committed to iterating in partnership with customers who have already begun developing MCP servers and experimenting with agent-based workflows.

FAQs

No, MCP Registries in Catalog is only available in Tech Preview via Konnect Labs.

This feature is built on top of Anthropic’s MCP Registry API specification, which is still rapidly evolving. Because the underlying standard continues to change, we cannot responsibly commit to GA timelines or long-term stability guarantees at this time.

We are actively iterating in partnership with customers who are exploring MCP-based agent architectures and will evaluate GA readiness as the specification matures.

If you’re creating a new MCP definition, create an MCP server. It’s the preferred object going forward, and is additive to the MCP registry’s server object. It captures remotes and packages like an MCP registry, but also supports the server’s tools, resources, and prompts. You can also link an MCP server to AI Gateway 2.0 as its source.

If you’re already using MCP registries, we recommend migrating to MCP server so you can use these additional capabilities. You can automate this by writing a script that iterates through each MCP registry and the MCP servers nested under it, then creates a new MCP server using the new object model for each one, including its remotes and packages. You won’t lose any information in the process, since the new object model is additive.

Help us make these docs great!

Kong Developer docs are open source. If you find these useful and want to make them better, contribute today!