AI | Anjana Data Documentación

This section contains the documentation on using AI assistants with Anjana Data Platform. It describes the components that Anjana Data SL makes available so that an AI assistant chosen by the organization can work with the platform, how they fit together and which responsibilities belong to each party.

Before using any of these components, read the Responsibility and human oversight section.

Anjana Data Platform approach

Anjana Data Platform is the Data & AI Governance Platform: it governs assets, their metadata, their lifecycle and access to them. To work with AI assistants, Anjana Data SL does not offer its own AI, but components that allow the AI each organization already uses to operate on the platform under the same governance rules.

Anjana Data SL does not provide AI models, the language model (LLM) is supplied by the AI client that the organization chooses and contracts, with the provider and terms it decides.
Anjana Data SL does not provide agents, assistants or chats. The MCP and the skills are installed in the organization's AI client; they do not run anything on their own.
Anjana Data SL does not analyze the data exchanged through the MCP or the skills: conversations, requests and results flow between the user, their AI client and their model provider, without passing through Anjana Data SL.

Section contents

Page	What it is	What it is for
MCP	Server that exposes the platform's governance operations to AI assistants through the open Model Context Protocol standard.	Discover and query assets, catalog and enrich metadata, submit for validation, handle access requests to DSAs and query the audit log in natural language.
Skills (coming soon)	Packages of instructions, knowledge and scripts that guide the AI assistant in specific tasks on the platform. They will be published in the Anjana Data community.	Automate platform configuration tasks and prepare Excel asset upload files, among others.

MCP and skills: how they differ

Both components are installed in the organization's AI client, but they serve different and complementary purposes:

	MCP	Skills
What it gives the assistant	A set of operations it can invoke on the platform.	Instructions and knowledge on how to carry out a specific task step by step.
Where it runs	Platform component, deployed in the Anjana Data Platform environment.	Files installed on the user's computer or AI client.
How it interacts with the platform	Through the platform APIs, with the identity and permissions of the authenticated user.	Depending on the skill: through the platform API with the user's credentials, or by generating local files that the user then uploads to the platform.
Governance controls	Assets are created in `DRAFT` and require human approval; all actions are recorded in the audit log.	Those of the platform for the channel used. Some changes, such as configuration changes, are applied immediately and do not go through an approval workflow.

The first planned skills cover two tasks:

Anjana Data Platform configuration, creation of users, organizational units, roles, permissions, metadata attributes, reference values, entity types, filters, notifications, translations and validation workflows, through the platform API.
Asset upload, filling in and editing the platform's Excel upload templates with the assets of a use case (datasets, business terms, processes, DSAs, relationships…). The skill only generates the files; the user performs the upload from the platform.

Responsibility and human oversight

Using AI assistants does not release the user from their responsibilities

Every action an AI assistant performs on Anjana Data Platform, through the MCP or through a skill, is executed on behalf of the user and with their permissions. The user remains responsible for those actions and must know the platform well enough to judge whether they are correct. Actions must always be carried out under human oversight.

This oversight is necessary because no component completely eliminates the risk of error:

Skills may contain errors, like any software. An incomplete instruction or an unforeseen case can produce an incorrect result.
The language model may deviate from the instructions, the restrictions of a skill are guidance for the assistant, not technical controls. The model may misinterpret them, skip them or act outside them, for example when trying to resolve an error on its own.
The assistant may misinterpret, it may choose an asset, environment or value different from the one the user had in mind, or present an imprecise conclusion as fact.
Some changes are immediate, configuration changes made through the API take effect at once and affect all users of the environment; not all of them can be undone.

Recommended guardrails

The organization and each user must establish their own controls. As a minimum, it is recommended to:

Test first in a non-production environment, and take to production only what has been validated.
Work with least privilege, connect with a user that has only the roles and permissions needed for the task, rather than a personal administration account.
Review each action before confirming it, and do not enable modes in the AI client that run operations without asking for approval.
Review the results, check the created or modified assets in the Data Portal and review the generated Excel files before uploading them.
Save the previous state, export or document the existing configuration before significant changes.
Protect credentials, use expiring session tokens or temporary passwords, and do not share them in conversations or files.
Review the audit log, periodically check the platform audit log to verify the actions performed.
Choose the AI provider according to internal policies, in particular regarding the processing and retention of the information the model receives.

The platform's own controls (role-based permissions, asset lifecycle, human approval and audit log) still apply, but they do not replace user oversight.

The most frequent questions about AI, privacy and determinism are answered in the FAQ section of MCP.

This section is intended for data and AI governance leads, functional and technical administrators, and security, architecture and compliance teams that evaluate or use AI assistants with Anjana Data Platform.