Press CTRL+K to open search

AI Destinations Overview

  • 6 minute(s) read
Prev Next

An AI destination is an AI assistant, agent, or other LLM client that reads data from Dataddo so you can ask questions about it in plain language. This can be:

DATADDO TIP

For the full list of available sources and destinations, see the full list of available connectors.

AI destinations work differently from other data destinations. Dataddo doesn't write data into the AI tool. Instead, it keeps the extracted data in SmartCache, and the AI client retrieves it on demand through the Data Access MCP Server.

DATA TO AI PLAN

On the Data to AI plan, you work with contexts and AI Models instead of sources, flows, and destinations. See Data to AI Plan for how to set them up.

How AI Destinations Work

  1. Sources (contexts on the Data to AI plan) extract data from your business applications on the schedule you set.
  2. SmartCache stores the latest extracted data and keeps it ready to query.
  3. The Data Access MCP server (https://headless.dataddo.com/mcp-data) serves the data through a semantic layer. The model asks in business terms (customers, invoices, deals), with filters, grouping, date grains, and predefined metrics. Dataddo never accepts SQL from the model.
  4. The AI client connects to the server once, and you sign in with your Dataddo account through OAuth. Every answer comes with metadata: technical metadata such as data freshness, and business metadata such as what each table and column means.

The tools the server exposes let the model discover data (list_flows, list_entities, describe_entity, search), query it (query), and check its freshness and limits (data_status, known_gaps). For the full tool list, authentication details, and rate limits, see Data Access MCP Server.

AI Destinations on Your Plan

The way you attach data to an AI client depends on your Dataddo plan:

Data Anywhere and Enterprise plans Data to AI plan
Your data comes from Sources (Sources page) Contexts (Context page)
The AI client is Destination (Destinations page) AI Model (AI Models page)
Data is connected to it by Flows (Flows page) Attached contexts, shown on the Governance & Lineage page

A context is data that Dataddo keeps ready to query. On both plans, you:

  1. Create an authorizer for the AI client.
  2. Attach the data.
  3. Add the Dataddo connector in the AI client.

For step-by-step instructions, see the articles dedicated to individual clients.

Data to AI Plan

On the Data to AI plan, you work with contexts and AI Models instead of sources, flows, and destinations.

Create a Context

A context holds the data from one of your business applications and keeps it ready to query.

  1. Go to the Context page and create a new context.
  2. Select the application and configure the data the same way you would create a data source.
DATADDO TIP

Exclude or hash personal data while you configure the context. See PII Exclusion and Hashing.

Activate an AI Model

  1. Go to the AI Models page and click Activate AI Model.
  2. Select your AI client.
  3. Follow the on-screen instructions to authorize the connection and connect the AI client to Dataddo.
  4. Attach one or more contexts to the AI Model.

For detailed instructions for each client, see the article for that client, e.g. Claude, ChatGPT, Cursor, Google Gemini, Mistral, Copilot, Grok, or Generic MCP.

Governance & Lineage

The Governance & Lineage page gives you an overview of which context is used in which AI Model. Use it to check what data each AI client can access.

Selecting the Optimal AI Destination

All AI destinations read the same data through the same server, so the choice mostly depends on where your team already works. This guide matches common use cases to the clients best suited for each, and shows when a different type of destination is a better fit.

Use Case Destination
Ad-hoc questions about business data in a chat An AI assistant such as Claude, ChatGPT, Le Chat, or Grok. Ask across sources such as ad platforms, analytics, and CRM in one conversation. On team and enterprise plans of these assistants, an admin usually adds the connector once, and each member signs in with their own Dataddo account.
Working with data while you code or in the terminal A coding assistant such as Claude Code, Cursor, GitHub Copilot in VS Code, or Gemini CLI.
Data in Microsoft 365 and Teams Microsoft Copilot. Add the Dataddo MCP server to an agent in Copilot Studio and publish the agent to Microsoft 365 Copilot and Teams.
Custom agents and other AI tools Generic MCP. Any client that supports Streamable HTTP and MCP OAuth (with PKCE and dynamic client registration or a Client ID Metadata Document) can connect with the endpoint URL alone.
RAG, fine-tuning, or AI that needs long history SmartCache holds recent extractions, not long time-series ranges. For long history, send data to a cloud data warehouse such as Snowflake, BigQuery, or Databricks, or to object storage such as Amazon S3, and point your AI pipeline at it.
Real-time operational AI (e.g. support chatbots) For near real-time data without loading your production database, use Real-Time CDC Database Replication to keep a replica your AI system queries.

For a detailed comparison of these patterns, see Data Delivery to AI & Agents.

Data Access and Security

  • Only attached data is exposed: Flows and contexts that aren't attached to an AI destination or AI Model aren't available to any AI client.
  • Access is read-only: The AI client can't change your sources, flows, contexts, or the data in them. The only tool that records anything is report_problem, which logs an answer that looks wrong.
  • Exclude or hash PII before it reaches the model: Use PII Exclusion and Hashing on the source or context so sensitive columns never reach the AI client.
  • Data must be covered by a semantic model to be queryable: The AI client can see a flow or context as soon as it is attached, but it can query it only after a semantic model covers it. list_flows reports this for each one.
  • Data appears after the first run: A newly attached flow or context holds no data until it runs for the first time.

Other AI Destination Operations

Duplicate/Clone an AI Destination

  1. Click on the Clone Settings button next to your AI destination.
  2. Proceed with the on-screen prompts.

Delete an AI Destination

To delete an AI destination:

  1. Go to the Destinations page. On the Data to AI plan, go to the AI Models page.
  2. Click the bin icon for the relevant AI destination.
  3. Type DELETE to confirm your action.

Deleting the AI destination doesn't disconnect the AI client from your Dataddo account. To revoke the client's access, also remove the Dataddo connector in the AI client.

Troubleshooting

Restart Destination

If an AI destination enters a broken state (marked red), you can attempt to restart it. If successful, the destination will start working again. If restarting fails, an error message will let you know where the issue is.

In most cases, you will need to reauthorize your service.

Connection Issues in the AI Client

If the AI client cannot reach your data, the problem is usually on the client side, not in Dataddo:

  • The connector asks you to connect again. The sign-in expired or was revoked. Reconnect the connector and sign in to Dataddo.
  • The client cannot connect. Check that the URL is exactly https://headless.dataddo.com/mcp-data, that you did not add an authorization header, and that the client supports the Streamable HTTP transport. SSE-only clients are not supported.
  • You see data from the wrong account. Disconnect the connector and sign in again with the right Dataddo account.
  • Stale data. Review the extraction schedule of the source or context. Dataddo serves the most recent data it has stored.

For client-specific issues, see the Troubleshooting section of the article for your client.

Related Articles

Troubleshooting

Error Reporting Best Practice Troubleshooting