MCP servers in Catalog

As teams build MCP servers for AI agents, those servers are often embedded directly in agent code, defined in local configuration files, or scattered across repositories, with no shared record of what already exists. Catalog gives you a single place to see every MCP server across your organization, so you don’t have to track each one down in a separate tool to know what it does or how to connect to it. An MCP server is an interface in Catalog that represents an MCP server your organization builds or consumes.

You can create an MCP server in Catalog in a few ways:

• Define it manually, filling out its details yourself.
• Paste an existing server’s JSON definition.
• Import it from AI Gateway 2.0, by linking an existing AI Gateway MCP server as the source.
• (Coming soon) Connect to a running MCP server, and have Konnect fetch its definition directly.

When an MCP server is linked to AI Gateway 2.0, the link indicates that AI Gateway is protecting and proxying that server. An MCP server that isn’t linked can still exist in Catalog as part of your organization’s inventory. The absence of a link can also help you identify MCP servers that aren’t yet protected by AI Gateway and could be candidates to govern with it.

The link is a snapshot, not a live sync: if you change the linked server’s configuration in AI Gateway, the MCP server in Catalog isn’t automatically updated to match.

1. In the Konnect sidebar, click Catalog.
2. From the New dropdown menu, select MCP server.
3. Under Data source, select one of the following:
   1. To fill out the server’s details yourself, select Define manually.
   2. To populate details from an existing server’s definition, select Paste JSON, then in the Version, capabilities, and access methods field, paste the definition. Only version is required.
   3. To create the server from an MCP interface already configured in AI Gateway 2.0, select Import from AI gateway, then:
      1. From the AI gateway dropdown menu, select your AI Gateway control plane.
      2. From the MCP server dropdown menu, select the server you want to import.
4. In the Display name field, enter a name for your MCP server, for example My MCP server.
5. In the Name field, enter a unique identifier for the MCP server, for example my-new-mcp-server. This must contain only lowercase letters, numbers, hyphens, and periods.
6. (Optional) In the Description field, describe the purpose of your MCP server.
7. (Optional) Click Add labels, and do the following:
   1. In the Key field, enter a label key.
   2. In the Value field, enter a label value.
   3. (Optional) Click Add another label to add more labels.
8. Click Next.
9. If you didn’t select Paste JSON, in the Version field, enter a version number for your MCP server, for example 1.0.0.
10. (Optional) Under Capabilities, click New capability, and select one of the following:
    • Tool: Outlines how your server takes actions.
      1. In the Name field, enter a unique identifier for the tool.
      2. In the Input schema field, enter the JSON schema defining the tool’s expected parameters.
      3. (Optional) Fill in Title, Description, Output schema, and Annotations.
      4. Click Add, then repeat to add more capabilities.
    • Resource: To share server context through documents, logs, and other data.
      1. In the URI field, enter a unique identifier for the resource, for example file://example.txt.
      2. In the Name field, enter a name for the resource.
      3. (Optional) Fill in Title, Description, mimeType, and Size (bytes).
      4. Click Add, then repeat to add more capabilities.
    • Prompt: Define a reusable workflow with preset inputs.
      1. In the Name field, enter a unique identifier for the prompt.
      2. (Optional) Fill in Title, Description, and a comma-separated list of Arguments, for example language, focus_area, severity_threshold.
      3. Click Add, then repeat to add more capabilities.
11. (Optional) Under Access methods, click New access method, and select one of the following:
    • Remote: Describes a hosted MCP server endpoint.
      1. In the Remote URL field, enter the server’s endpoint, for example https://example.com/mcp.
      2. From the Transport type dropdown menu, select a transport type.
      3. (Optional) In the Headers field, enter headers as a JSON array of objects, for example, to pass an authorization token:

         [
           {
             "name": "Authorization",
             "value": "Bearer ${token}"
           }
         ]

         Copied!
      4. Click Add, then repeat to add more access methods.
    • Package, to describe a run-it-yourself distribution option.
      1. In the Package identifier field, enter the package name or URL, for example @modelcontextprotocol/mcp-server.
      2. From the Transport type and Package type dropdown menus, select the applicable types.
      3. (Optional) Fill in Registry base URL and Package version. If you leave Package version blank, the package uses the server’s version.
      4. Click Add, then repeat to add more access methods.
12. Click Save.