Summary

Resolves a critical compatibility issue between Cursor and DeepSeek's V4 thinking-mode API, specifically the missing reasoning_content field required for tool-call reasoning. This proxy injects the necessary data, enabling DeepSeek models to function correctly within Cursor and other applications facing similar DeepSeek API integration challenges.

How It Works

The proxy intercepts Cursor requests, caches DeepSeek's reasoning_content, and injects it into tool-call requests that omit it, resolving API errors. It uses ngrok for public HTTPS URL access, bridging local development with Cursor. It also renders DeepSeek thinking tokens into collapsible Markdown and performs API normalization.

Quick Start & Requirements

• Installation: Via uv (curl -LsSf https://astral.sh/uv/install.sh | sh, then uv run deepseek-cursor-proxy) or conda (conda create -n dcp python=3.10 -y, conda activate dcp, pip install -e ., deepseek-cursor-proxy).
• Prerequisites: Python 3.10+, ngrok (install/authenticate), DeepSeek API key.
• Configuration: ~/.deepseek-cursor-proxy/config.yaml or flags (e.g., --no-ngrok).
• Cursor Integration: Add DeepSeek model, set Base URL to ngrok HTTPS endpoint (e.g., `https://your-subdomain