Instructing A.I. agent on Canvas API pagination

Question

I am building a Copilot agent to support faculty queries.  My agent is connecting to Canvas (the Beta site) API via an MCP connection in Microsoft Azure.  How do I instruct the agent on Canvas API pagination?  Can I just tell the agent to keep looping or do I have to ask for MCP modifications?  Any experience from the community is appreciated.

For example, to list students and faculty in an course, I often need more than 1 API call to get all of the data back.  I tried this instruction but it doesn't seem reliable.

* Call page 1.
* Check how many users are returned.
* Continue calling the API for page 2, page 3, page 4, etc.
* Stop when a page returns no users.
* Combine users from all pages before responding.

Thank you

jeremyharper221 · Answer

@Doug9

I would strongly recommend handling the pagination in the code layer before the AI agent ever sees it. LLMs are generally very unreliable at executing 'while-loop' instructions via natural language. Instead, you could write a lightweight Python MCP server (using a framework like fastmcp deployed via azure.funtions). Inside that server, use the canvasapi Python library—which handles Canvas's specific Link-header pagination smoothly and automatically behind the scenes. Your MCP tool just needs to route the agent's request to canvasapi, gather the complete list of users, and return the final output to the agent as a single JSON array. If you're not familiar with Python, you could even get an LLM to generate the basic code for you.

Robert · Answer

@Doug9

?per_page=100 is helpful, because the default is 10 records per page.

Also the rest of the pagination documentation should be useful for the model to interpret, or tell it to look for the last link.