Large SharePoint/OneDrive files fail in Copilot Studio tools and flows

(1) Share

Report

Posted on by VB-03020721-0

Hi Everyone,

I am facing a limitation while trying to access and process larger files from SharePoint / OneDrive inside a Copilot Studio agent.

Scenario

I have a Copilot Studio agent where I need to read documents from SharePoint / OneDrive and validate the content. The files can be PDF, DOCX, XLSX, PPTX, etc.

The current approach is:

Use Work IQ SharePoint / SharePoint tool to identify the required file.
Use Get file content using path connector action/tool to retrieve the file.
Return the file content back to the agent.

Issue observed

For smaller files, the connector/tool call may complete. But when the file is larger, the agent is not able to process the returned content properly.

I have tried multiple approaches:

Work IQ SharePoint tool
- It identifies the file/metadata.
- But when trying to read the file content, it shows that readSmallTextFile supports only text-based files under 5 MB.
- Example: one PDF file was around 8.07 MB, so it could not be retrieved/read directly.
SharePoint / OneDrive “Get file content using path” connector action as a tool
- The connector action completes successfully.
- Output returned is {File Content}.
- However, the file content is large and the agent fails while trying to process it.
Flow using Get file content using path
- I also tried calling the same connector action through a flow and returning the output to the agent.
- The flow/action completed successfully, but once the response is returned to the agent, agents fail with the payload is too large error.

Errors received

I received the following errors in different attempts:

Error Message: The output returned from the connector was too large to be handled by the agent. Try reducing its size by utilizing available connector filters or by limiting the number of configured action outputs.
Error Code: AsyncResponsePayloadTooLarge

Error Message: An error has occurred while processing the PDF. Please ensure the document is not empty and does not exceed the maximum number of pages allowed.
Error Code: InvalidPdf

Sorry, something went wrong.
Error code: SystemError.

Official limits I found

From Microsoft documentation, Copilot Studio web app limits mention:

Connector payload: 5 MB for public cloud plans.
The AsyncResponsePayloadTooLarge error means a real-time connector is returning a payload larger than the agent can handle.

This explains why even if SharePoint / Flow can technically retrieve the file, returning that raw file content to the agent fails.

My question

What is the recommended Microsoft-supported workaround for processing files larger than 5 MB inside Copilot Studio agents when the file must be read dynamically from SharePoint / OneDrive?

Specifically, I am looking for guidance on:

How to process SharePoint / OneDrive files larger than the 5 MB connector payload limit.
Whether there is a supported way to chunk file content and send it to the agent in multiple calls.

Expected behavior

I need the agent to dynamically read a selected SharePoint / OneDrive document and validate the contents, even when the document is larger than 5 MB.

Current blocker

The connector/tool can retrieve the file, but the response cannot be handled by the agent because:

The returned file content is binary.
The file/payload can exceed the Copilot Studio connector payload limit.
Large PDF/DOCX content cannot be passed back to the agent directly.
Returning file content through flow still hits the same “too large response” issue.

Has anyone implemented a reliable pattern for this scenario?

Any recommended architecture or workaround would be very helpful.

Categories:

General topics

I have the same question (0)

All responses (7)

Answers (0)

Sort by

Suggested answer

Haque 3,653 on at

Like
a
(2)

Report
Copy link

Link copied!
Hi @VB-03020721-0,

The core challenge you face is that large files (over 5 MB) and binary content cannot be passed directly back to the Copilot Studio agent due to payload size limits and the binary nature of the data.

Here are some workarounds:

Use Chunking or Pagination: Instead of returning the entire file content at once, split the document into smaller chunks (e.g., by pages or sections) within your Power Automate flow. Then, pass only the relevant chunk to the agent for processing or validation.

Pre-Process and Extract Text Outside the Agent: Use Power Automate or Azure Cognitive Services (e.g., Form Recognizer, OCR) to extract and preprocess the text content from the document before passing it to the agent. Store the extracted text in a temporary storage (Dataverse, Azure Blob Storage, or SharePoint) and have the agent query or retrieve only the processed text.

Use External Storage and Reference Passing: Instead of passing the file content, pass a reference (URL or ID) to the stored file or extracted content. The agent can then request or query the relevant parts dynamically.

Convert Binary to Base64 with Caution: While converting binary content to Base64 is possible, it increases payload size by about 33%, making it less suitable for large files.

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No
VB-03020721-0 2 on at

Like
a
(1)

Report
Copy link

Link copied!

Hi @Haque,
Thank you for the response and for sharing these workaround ideas. They are helpful, but in my scenario they do not fully solve the issue.

For chunking/pagination, the challenge is that when a flow is used in Copilot Studio, it starts with When an agent calls the flow and ends with Respond to the agent. So even if the flow retrieves the file using Get file content using path, the final full data still has to be returned to the agent (No way for chunking the file and return chunks recursively). If the response is large, it again fails with AsyncResponsePayloadTooLarge. Also, the SharePoint connector returns binary file content, so chunking raw binary does not help the agent read the document.

For preprocessing/OCR/external storage: our files are not fixed. They keep growing daily across different SharePoint/OneDrive folders and subfolders. Setting up OCR, temporary storage, indexing, and querying for every possible file would add a lot of overhead. Most files are normal Word documents or text-based PDFs, so OCR also feels like overengineering.

For passing only a URL/reference, the agent still needs a supported way to read that referenced file in smaller readable parts. Otherwise, the same connector payload/output limit applies.

For Base64, as mentioned, it increases payload size and is not practical for multiple file types like PDF, DOCX, XLSX, and PPTX.

Please feel free to correct me if any of my understanding is incorrect, or if there is a recommended implementation pattern that I have not considered.

Thanks again for the suggestions. I am still looking for a practical supported implementation pattern for this specific limitation.

Was this reply helpful? Yes No
Suggested answer

Vish WR 3,748 on at

Like
a
(2)

Report
Copy link

Link copied!

@VB-03020721-0

Chunking via flow, pre-processing with Azure Form Recognizer/OCR and storing extracted text externally, passing a URL reference instead of raw content, or Base64 encoding

The real problem: The user pushed back on all of these — in Copilot Studio flows, the entire response still has to come back via a single "Respond to the agent" step, so chunking doesn't help. Files are also dynamic and grow daily across many folders, making pre-processing/OCR pipelines impractical overhead for mostly normal Word docs and text PDFs.

Was this reply helpful? Yes No
Suggested answer

chiaraalina 2,425 Super User 2026 Season 1 on at

Like
a
(2)

Report
Copy link

Link copied!
Hi @VB-03020721-0

Microsoft’s recommended pattern is to process and chunk documents outside the agent, then let the agent query only relevant text snippets:

Index SharePoint/OneDrive documents in Azure AI Search

Use integrated chunking and vectorization

Connect Azure AI Search as a knowledge source in Copilot Studio

The agent retrieves only the top relevant chunks + citations, not the full file

This avoids payload limits and scales cleanly to large and numerous documents.

If you have a Microsoft 365 Copilot license and enable Work IQ, Copilot Studio can use Microsoft’s semantic index for SharePoint. In that case:

SharePoint files up to ~200 MB are supported

No raw file content is returned to the agent

The agent reasons over indexed content instead of connector payloads

This is the simplest option if your tenant already has M365 Copilot licensing.

Hope it helps!

Was this reply helpful? Yes No
Suggested answer

Sajeda_Sultana 189 on at

Like
a
(1)

Report
Copy link

Link copied!
Hi @VB-03020721-0,

You're actually on the right track. What you're seeing is a current limitation of Copilot Studio rather than an issue with your implementation.

Copilot Studio has a response size limit (approximately 5 MB) for actions and Power Automate flows. Because of this limit:

Large files such as PDFs, Word documents, and Excel files cannot be passed directly to the agent for processing.

Splitting the file into smaller chunks doesn't help, because the total response still needs to fit within the same limit.

Returning file content as binary or Base64 data will also hit the size restriction.

So, unfortunately, Copilot Studio isn't designed to process large document contents directly.

Recommended Approach

Instead of sending the file itself to Copilot Studio, send only the file location and process the document outside the agent.

A common architecture looks like this:

Copilot Studio

Sends the file path (SharePoint, OneDrive, etc.)
Sends the user's question

⬇️

Power Automate

Passes the file path and question to an Azure Function (or another backend service)

⬇️

Azure Function

Reads the file directly from storage
Extracts and analyzes the content
Optionally uses Azure OpenAI for summarization, validation, or Q&A
Returns only a small response back to Copilot Studio

⬇️

Copilot Studio

Receives the concise response
Displays the result to the user

This approach works well because the large file never needs to pass through Copilot Studio, and only the final answer is returned to the agent.

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Sajeda_Sultana

Was this reply helpful? Yes No
Suggested answer

Haque 3,653 on at

Like
a
(2)

Report
Copy link

Link copied!
Hi @VB-03020721-0,

Interestingly, you requirement and concern both are valid.

Unfortunately, there is no out-of-the-box, fully supported way for Copilot Studio agents to dynamically read large SharePoint/OneDrive files in smaller chunks directly from a URL or reference without hitting payload limits.

I agree with chiaraalina - That's the only recommended approach right now based on your requirement hitting recommended approaches that Sajeda_Sultana mentioned.

Plus: The recommendation remains from my side are:

Pre-extract and preprocess text content outside the agent, on demand, focusing on text-based files (Word, text PDFs).

Store extracted text or summaries in a structured data store (Dataverse, SharePoint list) that the agent can query in small pieces.

Use caching and incremental retrieval at the data store/query level, not at the flow response level.

Avoid passing raw binary or full file content back to the agent.

Reference:

Text extraction from file in Copilot Studio - How?

Copilot Studio quotas and limits

Pass files to agent flows, connectors, and tools

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No
Vish WR 3,748 on at

Like
a
(1)

Report
Copy link

Link copied!

@VB-03020721-0

wanted to check if you were able to resolve your issue ?

Please ✅ Does this answer your question if my post helped you solve your issue. This will help others find it more readily. It also closes the item. If the content was useful in other ways, please consider answering Yes to Was this reply helpful? or give it a Like ♥
Visit my blog My Tech Space LinkedIn

Was this reply helpful? Yes No