Integrate with AI [BETA]
This is a beta functionality. API may change, and LLM-generated responses can be inconsistent or inaccurate.
The large-scale adoption of large language models has permanently shifted the flow of work related to data governance. As a response to these rapid changes, and to enable a smoother, more seamless adoption of AI tools into your work, Dataedo offers the option of integrating AI models with your repository using MCP.
Why Integrate?
One question you might have is why bother integrating an AI service with your repository if you can use existing chatbots without any additional configuration. The answer lies in context. Non-integrated models will lack the fundamental awareness of the organization of your repository and the assets within it. Even if you were to very meticulously describe the local context, there are some dependencies you would probably miss. This lack of context can lead to hallucinations; the model would simply make a less-than-educated guess, filling in the gaps using assumptions as necessary.
Model integration, on the other hand, provides the model with controlled access to your data. Now, it can perform better queries and have access to necessary information, leading to better insights, even though you, as a user, will probably provide the same prompts to the AI chatbot.
You can even connect the integrated model with your Microsoft Teams environment, to create an interactive chat directly in Teams, which employees can quickly use to draw conclusions from the data documented in your repository.
What is MCP?
MCP stands for Model Context Protocol. It is an open standard, developed by Anthropic, for connecting AI agents to external data, systems and applications in a safe manner. Think of it as an API for your AI models that allows for controlled access to your data.
MCP is also what we use to enable the connection of AI models to Dataedo.
Dataedo's MCP server is a small service that sits between your AI tool and Dataedo. It is therefore a standalone tool, not an integrated part of Dataedo Portal. When you ask a question in your AI client, the MCP server calls Dataedo's API, which reads the answer from your metadata. The AI gets read-only access, and cannot modify the data stored in your repository.
What can your model do?
Currently, Dataedo's MCP integration supports the following actions:
- search for objects in your repository catalog by name, title or description
- get detailed information about tables, columns, databases, Data Lineage, etc.
- list tables and procedures in a specific database; the list can also be optionally filtered
How authentication works
In Hosted mode, AI clients sign users in through your Dataedo instance's own login page (using any login method enabled there, including SSO) via an OAuth application registered in Settings > OAuth applications. The Portal then issues short-lived, signed access tokens (plus refresh tokens) that the MCP server validates on every request. By default those tokens are signed with keys held only in the Portal's memory, so they do not survive a Portal restart. See Keep sign-ins across Portal restarts to avoid repeated logins. In Local mode, you sign in with a Personal Access Token generated from your Dataedo profile instead of an organization login. No matter the mode you choose, the model only ever sees what your Dataedo account is allowed to see.
