llama.cpp MCP Client Gives Your Local AI Real World Skills

By vramkickedin | | 2 min read
A robotic agent in a white suit displaying MCP client

A new update to llama.cpp introduces an MCP Client that brings tool use and agentic capabilities to local AI workflows. This client connects llama.cpp to external tools and data sources through the Model Context Protocol.

Developer allozaur spent over a month building this feature set, which turns a standard local inference setup into something more interactive. The update targets users who want their local models to actually do things rather than just generate text.

Expanded functionality for local models

  • Tool calls allow models to interact with external services and data.
  • Agentic loop logic with processing stats visible in the UI.
  • Resource browser with search and file tree navigation.
  • Prompt system with arguments, attachments, and picker tools.
  • Server selector with capability cards showing what each server offers.
  • CORS proxy built into the llama-server backend.

Professionals running local AI for sensitive work can now connect their models to internal tools without sending data to external APIs. The resource browser and attachment system makes it easier to feed documents and files directly into conversations, which helps with analysis tasks.

Developer notes and current MCP Client status

This pull request represents substantial architectural changes to the codebase. The developer created new components for collapsible content blocks, improved markdown rendering, and added better handling for code blocks with syntax highlighting. However, users should proceed with caution. As noted in the project update:

'Assume this is a work in progress, guys, so proceed only if you know what you're doing.'

The feature requires enabling a specific flag when running llama-server. Users need to run the command with `--webui-mcp-proxy` to activate the CORS proxy functionality. Several previous pull requests laid the groundwork for this release, covering architecture improvements and UI foundations.

Get llama.cpp MCP Client on GitHub.