--- title: MCP servers in Catalog description: An MCP server is an interface in Catalog that represents an MCP server your organization builds or consumes. Learn how to create and manage MCP servers. url: "/catalog/mcp-servers/" canonical_url: "/catalog/mcp-servers/" content_type: reference products: - Catalog - AI Gateway canonical: true works_on: - konnect --- # 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](/ai-gateway/), by linking an existing [AI Gateway MCP server](/ai-gateway/entities/ai-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. ## Create an MCP server ### Konnect API Send a POST request to the `/mcp-servers` endpoint to create an empty MCP server: ```bash MCP_ID=$(curl -X POST "https://us.api.konghq.com/v1/mcp-servers" \ --no-progress-meter --fail-with-body \ -H "Authorization: Bearer $KONNECT_TOKEN" \ --json '{ "name": "my-mcp", "display_name": "My MCP", "description": "Customer support triage MCP server" }' | jq -r ".id" ) ``` Add the MCP server's capabilities and access methods by sending a POST request to the `/mcp-servers/{mcpId}/versions` endpoint: ```bash curl -X POST "https://us.api.konghq.com/v1/mcp-servers/$MCP_ID/versions" \ --no-progress-meter --fail-with-body \ -H "Authorization: Bearer $KONNECT_TOKEN" \ --json '{ "version": "1.0.0", "tools": [ { "name": "get_forecast", "description": "Multi-day forecast for a location.", "input_schema": { "type": "object" } } ], "resources": [ { "name": "forecast-docs", "uri": "file://forecast-docs.txt", "description": "Reference documentation describing forecast data fields and units.", "mime_type": "text/plain" } ], "prompts": [ { "name": "summarize_forecast", "description": "Summarize the forecast for a location over a date range.", "arguments": [ { "name": "location", "description": "The location to summarize the forecast for.", "required": true } ] } ], "remotes": [ { "type": "streamable-http", "url": "https://mcp.example.com/weather" } ], "packages": [ { "registry": { "type": "npm" }, "identifier": "@example/weather-mcp-server", "version": "1.0.0", "transport": { "type": "stdio" } } ] }' ``` (Optional) Link the MCP server to a server on AI Gateway by sending a POST request to the `/mcp-servers/{mcpId}/implementations` endpoint: ```bash curl -X POST "https://us.api.konghq.com/v1/mcp-servers/$MCP_ID/implementations" \ --no-progress-meter --fail-with-body \ -H "Authorization: Bearer $KONNECT_TOKEN" \ --json '{ "implementation": { "type": "ai-gateway", "config": { "gateway_control_plane_id": "'$CONTROL_PLANE_ID'", "gateway_mcp_server_id": "'$GATEWAY_MCP_SERVER_ID'" } } }' ``` ### Konnect UI 1. In the Konnect sidebar, click **Catalog**. 1. From the **New** dropdown menu, select **MCP server**. 1. Under **Data source**, select one of the following: 1. To fill out the server's details yourself, select **Define manually**. 1. 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. 1. 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. 1. From the **MCP server** dropdown menu, select the server you want to import. 1. In the **Display name** field, enter a name for your MCP server, for example `My MCP server`. 1. 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. 1. (Optional) In the **Description** field, describe the purpose of your MCP server. 1. (Optional) Click **Add labels**, and do the following: 1. In the **Key** field, enter a label key. 1. In the **Value** field, enter a label value. 1. (Optional) Click **Add another label** to add more labels. 1. Click **Next**. 1. If you didn't select **Paste JSON**, in the **Version** field, enter a version number for your MCP server, for example `1.0.0`. 1. (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. 1. In the **Input schema** field, enter the JSON schema defining the tool's expected parameters. 1. (Optional) Fill in **Title**, **Description**, **Output schema**, and **Annotations**. 1. 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`. 1. In the **Name** field, enter a name for the resource. 1. (Optional) Fill in **Title**, **Description**, **mimeType**, and **Size (bytes)**. 1. 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. 1. (Optional) Fill in **Title**, **Description**, and a comma-separated list of **Arguments**, for example `language, focus_area, severity_threshold`. 1. Click **Add**, then repeat to add more capabilities. 1. (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`. 1. From the **Transport type** dropdown menu, select a transport type. 1. (Optional) In the **Headers** field, enter headers as a JSON array of objects, for example, to pass an authorization token: ```json [ { "name": "Authorization", "value": "Bearer ${token}" } ] ``` 1. 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`. 1. From the **Transport type** and **Package type** dropdown menus, select the applicable types. 1. (Optional) Fill in **Registry base URL** and **Package version**. If you leave **Package version** blank, the package uses the server's version. 1. Click **Add**, then repeat to add more access methods. 1. Click **Save**. ## FAQs - Should I create an MCP server or MCP registry in the new Catalog? If you're creating a new MCP definition, create an [MCP server](/catalog/mcp-servers/). 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](/catalog/mcp-servers/#create-an-mcp-server) 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. ## Related Resources - [Catalog](/catalog/) - [AI Gateway](/ai-gateway/) - [MCP registries (tech preview)](/catalog/mcp-registry/) - [AI Models](/catalog/ai-models/)