Skip to main content

MCP Marketplace Validation

This guide walks through integrating, validating, and launching external MCP servers on the Databricks Marketplace; from first technical checks to a live, discoverable listing so customers can use these tools inside their Databricks AI workflows.

Prerequisites​

Before starting MCP Marketplace validation:

Overview: Four phases to launch​

PhaseDescription
Test ConnectionValidate technical connectivity by enabling the MCP preview and verifying tool functionality in AI Playground
Integration ReviewComplete a 30-minute live review with Databricks Partner Engineering
Create and Publish ListingCreate provider profile, configure MCP listing, and publish to the Marketplace
Test Published ListingInstall and test the published listing end-to-end
Launch and AmplifyMarketplace launch complete; update partner documentation to describe MCP usage on Databricks

Step 1: Test the MCP connection​

Before the Integration Review, validate that the MCP server works with Databricks by testing in a workspace.

Follow the Databricks documentation to:

  1. Enable the MCP preview - How to manage previews
  2. Create a Unity Catalog MCP Connection - Use external MCP servers
  3. Test in AI Playground - Test MCP servers with AI Playground

Step 2: Integration Review​

The Integration Review is a 30-minute live session with Databricks Partner Engineering to validate your MCP integration.

What happens during the review​

  1. Live validation: Demonstrate your MCP server connection and tool invocation
  2. Technical verification: Confirm end-to-end functionality with Databricks-hosted LLMs
  3. Approval: If successful, your integration is marked as validated

After validation​

Once the integration passes validation:

  1. The Partner Engineer will internally allowlist the workspace for MCP Marketplace publishing
  2. Once allowlisting is complete, proceed to create the Marketplace listing

Information to prepare​

Have the following ready for your listing:

ItemDescription
Short descriptionA brief overview of what the MCP server does and the problems it helps customers solve (up to 160 characters)
Tools exposedA list of tools available through your MCP server, with descriptions. Include planned roadmap tools
Joint value propositionHow joint customers will use your MCP integration with Databricks, including key use cases

Step 3: Create and publish your listing​

After your workspace is allowlisted, create your MCP Marketplace listing.

Create a Provider Profile and MCP Listing​

Follow the Databricks documentation to create the provider profile and listing:

For MCP server authentication options and configuration fields, see the MCP server listing fields section in the Databricks documentation.

MCP listing configuration guidance​

General:

  • Listing name: Use fewer than 100 characters, title case (capitalize primary words)
  • Short description: Up to 160 characters, sentence case, appears in search results

Attributes and details:

  • Categories: Select up to five categories that consumers can use to filter listings; categories also appear as tags on listing tiles and detail pages
  • Description: Detailed description of the MCP server and tools being offered; supports basic Markdown formatting (bold, italics, bullets, numbered lists)
  • Documentation: URL linking to documentation that helps consumers understand and use the MCP server (e.g., API reference, getting started guide)
  • Privacy policy: URL to privacy policy; must be publicly accessible and require no login

Access options:

  • Instantly accessible: Consumers gain access directly from the Marketplace without approval (recommended for most MCP servers)
  • Require approval: Use when a business agreement is required before providing access

OAuth configuration:

  • Token endpoint: Usually in the format https://authorization-server.com/oauth/token
  • Authorization endpoint: Usually in the format https://authorization-server.com/oauth/authorize
  • Credential exchange method:
    • header_and_body: Default, places credentials in both header and body
    • body_only: Credentials only in request body (no authorization header)
    • header_only: Credentials only in authorization header (use for providers like Okta)

Publish the listing​

  1. Go to Marketplace → Provider Console
  2. Click Publish listing
  3. The listing will appear in the public MCP Marketplace within a few minutes

Step 4: Test the published listing​

After publishing, test the live listing end-to-end to verify the consumer experience works correctly. This creates a new Unity Catalog connection (separate from the one used in Step 1) to simulate what customers will experience.

Install from Marketplace​

Follow the consumer installation process documented at Get access to external MCP servers:

  1. Open the Databricks Marketplace and search for the MCP listing
  2. Click Get or Install on the listing
  3. Enter the connection name and authentication credentials
  4. Complete the installation

Verify and test​

  1. Verify MCP preview is enabled - How to manage previews
  2. Test in AI Playground - Test MCP servers with AI Playground

Once testing is complete, the listing is ready for customer use.

What's next: Launch and amplify​

Once the listing is live, the Marketplace launch is complete. Partners should:

  • Update partner documentation to describe how customers can use the MCP server on Databricks
  • Share the listing URL on websites, blogs, and social channels

Resources​