
Hi,
I have a Copilot Studio topic that uploads:
The topic calls a Power Automate flow using When an agent calls the flow.
The flow works in Copilot Studio Test, but fails when run through the actual agent chat.
In the agent chat, the user gets Connect to continue prompts for:
The flow then fails at the AI Builder OCR step:
OCR_Extract_Text
Error:
The base 64 binary content is missing for input single
I have already checked/tried:
contentBytes + nameWhat makes this confusing is:
Because of that, the current suspicion is that this is not a PDF problem or a pure flow logic problem, but a runtime connection/authentication issue when the agent calls the flow. More specifically, the AI Builder connection does not seem to be fully established before the OCR step executes, which causes OCR to receive missing/null content and fail.
Main questions:
If anyone has seen this before, I would appreciate guidance.
Hi! You've done excellent diagnostics. The issue you're describing is a well-known connection configuration pattern — here's the explanation and fix.
In the Copilot Studio Test pane, flows run under your maker credentials, so all connector connections are pre-authenticated. In the published agent chat, flows run in the context of the end user — and unless connections are configured as "owner" connections, end users are prompted to authenticate each connector at runtime. This is the root cause of the "Connect to continue" prompt appearing mid-flow.
Per Use agent flows with your agent – Microsoft Learn, Power Automate flows called from Copilot Studio agents support connection references that can be set to use the flow owner's connection rather than requiring end-user authentication.
Consider migrating the flow to an Agent Flow using the "When an agent calls the flow" trigger — these are optimised to avoid user-facing connection prompts. Per Microsoft Learn, agent flows provide a seamless maker experience and avoid runtime auth prompts.
Found this helpful? Please mark ✅ "Does this answer your question?" so others searching for the same issue can find it quickly. A 👍 on "Was this reply helpful?" or a ♥ Like is also much appreciated!
Raghav Mishra — LinkedIn | PowerAI Labs