Connect your AI chats

Use FastBI from an AI chat that supports remote MCP (Model Context Protocol) servers. Your AI client can discover connected data, ask questions, and retrieve reports using the access you approve.

Connect with OAuth

1. In FastBI, open User settings → Connect your AI with MCP and copy the MCP server URL. For FastBI’s production service, it is https://fastbi.app/api/mcp/v1.
2. In your AI client’s integrations or MCP settings, add a remote server with this URL. Choose Streamable HTTP if a transport is requested, and OAuth if an authentication method is requested. Exact labels depend on the client.
3. Start the client’s connection action. It opens FastBI for sign-in using Google, Microsoft, or your email and password.
4. On Connect …, check the app name and Return address, choose the Organization, and review Allow access to. Approve only the features you need, then select Allow access. Cancel declines the connection.
5. Return to your AI chat and ask it to list the available FastBI capabilities. If the client requires tools to be enabled for each chat, enable FastBI there too.

Signing in identifies you; the separate approval grants the AI app access to the selected organization. Connecting does not grant access to every organization you belong to.

Copy the MCP server URL and choose which FastBI features your AI client can use.

Choose the allowed features

• Catalog lets the client browse connected data and schemas.
• Reports lets it read reports and run saved widgets. This is a useful starting point for repeated metric retrieval.
• Ad hoc analysis lets it ask questions and run analysis.
• SQL Workbench lets it plan queries, run SQL, and read results.

Ad hoc analysis and SQL require an Analyst role. Access remains subject to your active workspace membership and permissions. Requested data is sent to the AI client, and analysis and queries use your normal FastBI credits.

Ask a one-off question

Tell the AI chat to use FastBI and specify the source, metric, dates, timezone, and exclusions. For example, adapt this request to your data:

Use FastBI to show net revenue by week for the last four complete weeks in USD. Use our orders source, exclude cancelled orders, and use the workspace’s reporting timezone. Include a chart, the date coverage, and the sources used.

The client can use a matching saved report or start an AdHoc analysis. Answer any clarification requests, wait for the run to finish, and check the returned filters and totals. Continue in the same conversation for follow-up questions. Chart display depends on the client; open FastBI to inspect a result when the client cannot display it.

Retrieve data periodically

Set the schedule in an AI client or automation that supports authenticated MCP calls during scheduled runs. Adding the FastBI connection alone does not create a schedule. A client without this capability can still make one-off requests.

1. Create a saved report with the metric and filters you want to repeat. Confirm the report and widget before scheduling it.
2. Tell the client to run that widget for fresh results, rather than simply repeat a previously fetched answer. Define the period, timezone, frequency, and destination in the client’s scheduler.
3. Run it once manually and check the metric, date coverage, and data freshness. Then enable the schedule in the client.

An example schedule instruction is:

Every weekday at 09:00 Asia/Bangkok, use FastBI to run the Daily sales report’s Revenue widget for the previous completed day. Return revenue, order count, and the comparison with the preceding day. Include the period and when the source data was last refreshed.

Adapt report and widget names to your workspace. Whether this instruction creates a task depends on your client’s scheduling features; confirm the saved schedule there.

FastBI Jobs synchronize selected source data into the Lake. They refresh the data; the AI client’s schedule retrieves it. Arrange the retrieval after the required sync completes, and check coverage before treating a result as current. Repeated analyses and queries can consume credits on each run.

How OAuth works

The client discovers FastBI’s authorization service, sends you to FastBI to sign in and consent, then receives a short-lived authorization code at its return address. It exchanges that code using PKCE for access and refresh tokens. You do not paste your FastBI password into the AI client.

Access tokens last up to one hour. A compatible client refreshes them automatically within the 30-day authorization period; after that, reconnect and approve access again. Refresh does not extend the original grant indefinitely. Scheduled retrieval also needs valid authorization when each run starts.

To disconnect, open User settings → Connect your AI with MCP → Connected AI apps and select Revoke for the app. This stops future access; it does not erase data already received by that client.

Clients that require a manual token

Under Manual access tokens, enter a connection name, select Allowed features, and choose Create access token. Copy the token when it is shown and put it in the client’s bearer-token credential field for the same MCP URL. Keep it out of chat messages and shared configuration. Tokens created here expire after 30 days and can be revoked in User settings.

If a connection stops working, check its expiry, the selected organization, approved features, and your workspace role. If results are stale, check the report’s period and the source’s latest successful sync before rerunning.

To bring an external MCP data source into FastBI instead, see MCP integrations.