Question about MCP Resources vs Tools for CSV Graph Plotting

[Dhruv-mak]

Hi team,

I'm new to fast-agent and may be missing something obvious. I'm trying to understand the best approach for my use case:

I understand what MCP tools do, but I'm still confused about resources. I've noticed resources are typically used within tools, but I have a specific question:

If I have a function that plots graphs from CSV input, should I be using MCP resources? If yes, how would I configure an agent that has access to this resource and returns the generated image back to me?

Currently, I'm working with code similar to:

Any guidance on best practices for this scenario would be greatly appreciated!

Thank you.

[evalstate]

Hi @Dhruv-mak -- hope you don't mind, but I'm going to transfer this to discussions.

[Dhruv-mak]

No worries @evalstate, I should've created this post on discussion from the start. Thanks for moving it to discussion.

[evalstate]

Hi @Dhruv-mak, thanks for dropping by!

Have you tried the data-analysis example bundled with fast-agent which demonstrates producing graphs/charts from CSV data? Check it out if not.

Brief reply, and may well follow up with a bit more but a couple of points about MCP first:

• Tools are the only MCP type that get automatically added to the conversation context by the Host application. This article contains a little more info on parts of that: https://llmindset.co.uk/posts/2025/01/mcp-files-resources-part1/. The other thing that is recommended to automatically be added are "instructions" if the MCP Server and Host support them.
• Resources are a good way to expose content to the Host that may have been generated by the MCP Server, but not suitable for inclusion in a tool call result. Imagine a custom format file which can't be tokenized by the LLM. That could be exposed as a Resource after creation.
• For a Server that supports Sampling it is possible to dynamically generate content - but it doesn't sound like this would be an appropriate use-case (depending on where the CSV file is located/how it is being processed).
• Roots provide a configurable way for the Host and MCP Server to share content - and ultimately would be easy for the User to see.

When discussing Resources that are used within tools we are typically talking about the EmbeddedResource type - which is essentially an attachment to the message - rather than an MCP Resource which is a primitive offered by the Server (although of course there is overlap).

[edit] -- to add - to best help answer the question it's useful to know where the CSV file is in your scenario, and what the LLM's job is (e.g. calling a program or figuring out the data shapes...)