Feature Request: Do you plan to include support for MCP?

[ycFreddy]

Prerequisites

• I am running the latest code. Mention the version if possible as well.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

Support for MCP

Motivation

Hello,

Do you plan to include support for MCP?

That would be great to have MCP.

Possible Implementation

No response

[Marxist-Leninist]

MCP support works with any model that is intelligent enough to use it. Try an 8B model: https://huggingface.co/prism-ml/Bonsai-8B-gguf

If it doesn’t work with that one, it likely won’t work with others, as they won’t be intelligent enough. Tool use and MCP depend entirely on the model’s intelligence

[ycFreddy]

It works on llama.ccp official but not on your fork.

It works but it's slow because the official llama.cpp runs 1-bit models on CPU.

[khosravipasha]

Which branch did you try on?
Probably best wait for our changes to be merged with main llama.cpp (Metal backend already merged, CUDA and x86 and few other backends pending)

[Marxist-Leninist]

If you mean MCP support in the llama.cpp CLI/runtime itself, that’s separate from whether the model is capable enough to use tools once MCP is available. My earlier comment was about model capability, but it sounds like you’re asking about fork-level MCP integration.

[khosravipasha]

Yeah our branch was a bit older maybe MCP support was added to llama.cpp in the meantime.
Most backends should be merged in main llama.cpp by sometime next week so closing this for now.

[Marxist-Leninist]

f6235a4 guess we could sync this with our current local copy probably pretty harmless and will save others headache rather then waiting for upstream stuff

[khosravipasha]

@Marxist-Leninist I think the newly updated prism branch should have this right? And llama.cpp has our main backends merged now (Except x86 optimization remaining)