MCP 101

Calling a tool

0. Make sure that nothing is listening on ports 8000 and 8080. Open 3 generously sized terminals on your screen.
1. Download a sensible model. Qwen 3.5 4B is sensible.
2. Compile fresh llama.cpp:

   git clone https://github.com/ggml-org/llama.cpp && cd llama.cpp
   cmake -B build && cmake --build build --config Release -j 6

3. Launch the llama in terminal #1:

   ./llama-server -m ~/Downloads/Qwen3.5-4B-Q8_0.gguf --ctx-size 4096 --temp 1.0 --top-p 0.95 --top-k 20 --min-p 0.00 --verbose --webui-mcp-proxy

4. Clone this repository:

   https://github.com/behavioral-ds/mcp-example && cd mcp-example

5. Install deps: poetry install && poetry shell
6. Launch MCP in terminal #2: python mcp_serve.py
7. Execute the Agentic Call™ in terminal #3: python call.py
8. Observe the dance between LLM <-> Inference engine <-> MCP <-> Client.

Using MCP prompts

9. Open llama web UI at http://localhost:8080/, go to settings and add a new MCP server:

10. Select "MCP prompt" when drafting a new message:
    

11. That's your @mcp.prompt() parsed into UI element, click it:
    

12. ...and supply some meaningful content:
    

13. Then click "Use prompt" and rejoice: