IMPORTANT: This is a simple application for testing and developing the OKFN MCP server. Prototyping and quick changes are expected and code is mostly disposable.
A lightweight web chat interface that connects users to an AI provider (DeepSeek, OpenAI, Ollama, or any OpenAI-compatible API) while giving the AI access to tools served by an MCP server.
The AI answers questions exclusively using MCP tool data, not its own knowledge.
Zero frontend frameworks — plain HTML/JS/CSS. Backend is a minimal Flask app with no extra dependencies beyond httpx.
This tools was developed to work with, OKFN MCP server but should work with any MCP server that implements the spec.
Recommended to use uv as a package manager.
# Create a virtual environment and install dependencies
uv sync
# Activate the virtual environment
source .venv/bin/activate
# Configure your settings (choose one):
# 1. Edit local_settings.py with your AI_API_KEY and other overrides
# 2. Or set environment variables: AI_API_KEY, AI_BASE_URL, AI_MODEL, MCP_URL
# Run the server (port will be :8064)
python app.py
# Debug mode (port will be :5000)
flask run --debugThe chat UI is available at http://127.0.0.1:8064.
To know exactly what's happening internally, see the debug folder.