Most servers give the agent more to read; this one gives it somewhere to ask. The call blocks on a desktop window, and whatever you type — with the files and images you attach — comes back as the tool result. One setup step is easy to miss: the continue-chat tag only means anything once you paste the contents of rule_for_ai_EN.txt or rule_for_ai_VI.txt into your agent's custom instructions.
A Python server whose single tool, ai_interaction, opens a PyQt5 window on your own machine so the agent can put a question to you mid-task instead of guessing.
- A popup you type into, with your text handed back to the agent line breaks and all
- Files and folders picked from a workspace tree, attached to the reply with their relative paths
- Images dragged in, pasted with Ctrl+V or picked from a file dialog, converted to base64 for the model
- Several images at once, each with a preview you can click to enlarge or remove individually
- A continue-chat tag in the output that tells the agent whether to call the tool again when you are done
- English and Vietnamese interface, with the choice remembered between sessions
Python 3.8 or higher with the PyQt5 dependencies from requirements.txt, and the absolute path to mcp_server.py in your client config. Images stay on your disk in the user_images directory — nothing is uploaded.