Remote MCP with large files
Hi folks,
I’m currently trying to extend my existing MCP server with the ability to analyze large files, but I’m running into some issues.
The files I want to analyze range from around 100 KB to 1 GB. The main problem is that my server needs to run remotely, so simply passing a local file path and having the MCP server load the file does not work in my case.
I brainstormed this with Claude and ChatGPT and came up with a chunked upload approach, where the file is split into several chunks and uploaded to the MCP server. Each chunk would be sent as a separate tool call.
After several iterations, I still haven’t managed to get this approach working reliably, and I’ve burned through a lot of tokens during test uploads. I’m wondering whether I’m the only one facing this problem, or if there is an established concept or design pattern for solving it.
Has anyone else run into this issue and maybe found a good solution?