MCP Installation

This page describes how to set up the Anjana Data Platform MCP: what the organization must prepare on the platform and the network, how the AI client is configured and how each user authenticates. To understand what the MCP is and how it works, see MCP first.

Process overview

Installation has two parts with different owners. The first is done only once per environment; the second, once per user and AI client.

Prerequisites

• Platform version, Anjana Data Platform 26.1 or later.

• MCP-compatible AI client, supporting remote MCP servers over HTTP with OAuth authentication. For example Claude (Claude Desktop, claude.ai or Claude Code), Visual Studio Code with GitHub Copilot or Cursor.

• Platform user, with access to the platform and the roles required for the operations to be performed.

• Connectivity, the computer running the AI client must reach the environment's public URL over HTTPS, just as it reaches https://<url_entorno>/login to access the Data Portal.

Step 1. Enable the MCP on the platform

The MCP is distributed as one more platform component (adp-mcp). How it is enabled depends on the deployment type:

Paths that must be reachable

The AI client must be able to reach the following paths through the environment's public domain. If there is a reverse proxy, load balancer or WAF in front of the platform, they must be allowed and forwarded to the MCP service:

Step 2. Configure the AI client

Configuration consists of declaring the MCP server in the AI client pointing to the environment URL. There is no need to configure tokens, API keys or passwords: authentication takes place in the next step.

Generic configuration

Most clients accept a JSON block like the following, where <url_entorno> is replaced with the domain of the Anjana Data Platform environment:

{
  "mcpServers": {
    "anjana-data": {
      "type": "http",
      "url": "https://<url_entorno>/mcp"
    }
  }
}

Examples by client

• Claude Code, from the terminal: claude mcp add --transport http anjana-data https://<url_entorno>/mcp

• Claude Desktop and claude.ai, add a custom connector from the connector settings, entering the URL https://<url_entorno>/mcp.

• Visual Studio Code, in the workspace or user mcp.json file, with the servers key instead of mcpServers.

• Other clients, see their documentation on remote MCP servers over HTTP.

The exact location of these options may vary between versions of each client.

Step 3. Authenticate

The first time the AI client connects to the MCP server, it starts the OAuth 2.0 authentication flow (Authorization Code with PKCE):

1. The client opens the Anjana Data Platform login screen in the browser. In some clients the server must be activated first; for example, in Claude Code with the /mcp command.

2. The user signs in with their platform account, with the same username and password as in the Data Portal or with the SSO configured in the organization.

3. The browser returns control to the AI client, which is now connected on behalf of that user.

The user who signs in determines what the assistant can do. The MCP applies exactly the same roles and permissions as the platform, managed by Zeus. If a user has several profiles, they should consciously choose which one to connect with.

Step 4. Verify the installation

Once the session is connected, simply ask the assistant to describe the MCP. For example:

Connect to the Anjana MCP and list all its capabilities and the roles I am authenticated with.

The answer should confirm the connection, list the available operations (search, details, lineage, cataloging, relationships, DSAs and audit) and show the user's roles. If the roles are not the expected ones, sign out and authenticate with the right user.

Troubleshooting

This page is intended for technical administrators who enable the MCP on the platform, security and network teams that must expose the service, and users who configure their AI client to work with Anjana Data Platform.