MCP Server

The Delphix MCP server enables AI assistants to interact with Data Control Tower (DCT) through natural language using the Model Context Protocol (MCP). Supported AI clients include Claude Desktop, Cursor, and VS Code Copilot.

After configuring the MCP Server, you can ask questions about your test data management environment.

The following example prompts demonstrate how natural language can be used to perform common operations in DCT:

• I need to provision an Oracle and PostgreSQL VDB and place them in the same VDB Group.

• Add the Team: Alpha tag to all VDBs that were provisioned in the last hour.

• Run a refresh from parent against all VDBs that haven't been refreshed in the last week.

• List the PostgreSQL VDBs that have not been refreshed in the last 30 days.

• How much storage is Delphix saving me in TBs? Dollars?

• Which top five compliance jobs are most likely to fail? Why?

• List all VDBS that occupy more space than their parent dataset.

• Search for dSources with the largest snapshot sizes.

In dynamic mode, example workflows are not limited to a curated tool list. Any operation available through the DCT API—including Continuous Data, Continuous Compliance, Synthetic Data, PaaS, reporting, and management operations can be requested using natural language.

Use this quick start guide to configure the MCP server. For comprehensive instructions, see the MCP Server repository.

Prerequisites

• Data Control Tower (DCT) instance with API access.

• Personal DCT API Key

  • See Select a toolset for your persona for information about the permissions associated with each persona. You do not need the full set of permissions to run a given toolset.

• MCP-compatible client. For example, Claude Desktop, Windsurf, or Cursor.

• Bring Your Own LLM (BYO-LLM) configured via the MCP-compatible client.

  • The Delphix MCP Server does not interface with the DCT internal LLM.

• Python 3.11+ or Bash for local server execution.

Installation

Refer to the installation instructions in the MCP Server GitHub repository.

Configure an MCP client - quick start

Dynamic mode is the default toolset. You do not need to set DCT_TOOLSET unless you want to use a persona toolset. The following configuration includes DCT_TOOLSET for clarity. After you change DCT_TOOLSET, restart or reconnect the MCP server in your client to apply the new configuration.

{
  "mcpServers": {
    "delphix-dct": {
      "command": "uvx",
      "args": [
        "--from", "git+https://github.com/delphix/dxi-mcp-server.git",
        "dct-mcp-server"
      ],
      "env": {
        "DCT_API_KEY": "<your-dct-api-key>",
        "DCT_BASE_URL": "https://<your-dct-host>",
        "DCT_TOOLSET": "dynamic",
        "DCT_VERIFY_SSL": "false"
      }
    }
  }
}

Select a toolset for your persona

The MCP server uses a configured toolset to determine which tools are available to the client. The following table lists the supported toolsets.

Note: The auto option is not supported. Setting DCT_TOOLSET to auto raises a ValueError in get_configured_toolset().

Dynamic mode

Dynamic mode is the default toolset. The server runs in dynamic mode unless DCT_TOOLSET is set to a persona toolset. When dynamic mode is active, the server registers two tools: discovery and execute.

Dynamic mode provides full API access based on the DCT OpenAPI specification. In dynamic mode:

• Endpoints across DCT products, including Continuous Data, Continuous Compliance, Synthetic Data, PaaS, reporting, and management, are accessible because the tools are generated from the OpenAPI specification rather than a predefined list.

• Changes to endpoints do not require updates to the MCP Server code. When DCT adds or modifies an endpoint, it becomes available through discovery and execute after the cached specification is refreshed.

• Dynamic mode removes the need for separate domain-specific CRUD integrations. Endpoint changes are handled through the specification, without requiring MCP Server updates.

Coverage limitation

Coverage is limited to what is defined in the hosted DCT OpenAPI specification. Any functionality not included in the specification is not available in dynamic mode. New domains, such as Synthetic Data Generation, must be included in the specification to be accessible through MCP.

Discovery tool

The discovery tool combines the previously separate List and Get tools into a single tool for browsing and inspecting operations.

Use the discovery tool to browse the DCT API surface and retrieve operation details from a locally cached copy of the DCT OpenAPI specification.

The tool supports the following actions:

• list_tags: Lists every DCT domain (tag) with its operation count. For example, VDBs, dSources, Environments, Snapshots, and Synthetic.

• list_operations: Lists operations that you can filter by tag, HTTP method, or keyword, with support for pagination.

• get_operation_schema: Retrieves the fully resolved schema for a specific operation, including parameters, request body, responses, and whether the operation requires a confirmation gate.

Execute tool

The execute tool acts as a DCT API dispatcher that processes requests using the provided path and HTTP method. The execute tool performs the following steps:

1. Substitutes {paramName} placeholders in the path.

2. Looks up the operation in the cached specification.

3. Validates required parameters against the specification.

4. Applies the confirmation gate for destructive operations. The first call returns confirmation_required. To proceed, call the operation again with confirmed=true.

5. Dispatches the call to DCT and returns the result.

OpenAPI specification download and cache

At startup, the server downloads the OpenAPI specification from the configured DCT instance, validates it, and caches it locally on disk. The discovery and execute tools use this cached specification.

The specification is the single source of truth, and the tools reflect the DCT version in use.

Asynchronous operations

Many DCT mutating operations are asynchronous and they return a job that must be polled to determine success.

In dynamic mode, the assistant uses the discovery and execute tools to locate the job-status endpoint, poll for completion, and report the terminal outcome. This removes the need to identify the correct endpoint or interpret job state manually.

For example, you can instruct the assistant to “poll the job if it is async and let me know if it succeeds.”

Dynamic mode authorization

Authorization is determined by the DCT API key. No additional per-tool RBAC layer is applied.

The execute tool invokes operations permitted by the API key.

Configure key environment variables

Videos

• General overview of the MCP Server

• Dynamic mode

• Claude App configuration and sample usage

• Visual Studio configuration and sample usage

Limitations

• For dynamic (default) mode, the MCP server is not restricted to predefined toolsets and instead reflects the full scope of the hosted DCT OpenAPI specification.

• For persona toolsets, the Delphix MCP Server is limited to Continuous Data and general platform create, read, update, and delete (CRUD) operations. Limited Continuous Compliance information, such as job execution and state, is also available for monitoring.

• The Delphix MCP Server is not part of the Data Control Tower deployment package. It requires a separate installation and configuration.

• The Delphix MCP Server is a preview release and must not be used in production.

Support

The Delphix MCP Server is an open-source community project provided under the MIT License.

If you make modifications or improvements to the source code, submit a pull request for review.

For any questions, bugs, or feature requests, create an issue. Delphix Support does not assist with open-source community projects.