6.1 KiB

Raw Blame History

deepseek-cursor-proxy

Compatibility proxy connecting Cursor to DeepSeek thinking models (deepseek-v4-pro and deepseek-v4-flash).

What It Does

✅ Caches DeepSeek reasoning_content from regular and streamed responses, then restores it on later tool-call turns when Cursor omits it. If the exact original reasoning is unavailable, the proxy fails closed instead of sending a fake placeholder. See DeepSeek docs for more details.
✅ Mirrors streamed reasoning_content into Cursor-visible <think>...</think> text so that thinking tokens are shown in Cursor's UI. For BYOK/proxy mode, Cursor renders this as normal text, not as a native collapsible thinking block.
✅ Starts an ngrok tunnel so Cursor can reach the local proxy through a public HTTPS URL.
✅ Provides other compatibility fixes to make DeepSeek models run well in Cursor.

Why This Exists

This repository fixes the following Cursor + DeepSeek tool-call error with thinking mode enabled:

⚠️ Connection Error
Provider returned error:
{
  "error": {
    "message": "The reasoning_content in the thinking mode must be passed back to the API.",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_request_error"
  }
}

Usage

Step 1: Set Up ngrok

Cursor blocks non-public API URLs such as localhost, so the proxy needs a public HTTPS URL. ngrok can expose the local proxy to Cursor without opening router ports. Alternatively, you may use Cloudflare Tunnel.

Create an ngrok account, then visit ngrok's Dashboard: https://dashboard.ngrok.com

Then, install and authenticate ngrok once:

brew install ngrok
ngrok config add-authtoken <your-ngrok-token>

Step 2: Add Cursor Custom Model

In Cursor, add the DeepSeek custom model and point it at this proxy:

Model: deepseek-v4-pro
API Key: your DeepSeek API key
Base URL: your ngrok HTTPS URL with the /v1 API version path

The proxy respects the DeepSeek model name Cursor sends, such as deepseek-v4-pro or deepseek-v4-flash. The model field in config.yaml is only the fallback used when a request does not include a model.

For example, if ngrok dashboard shows https://example.ngrok-free.app, use:

https://example.ngrok-free.app/v1

Note: you can toggle the custom API on and off with:

macOS: Cmd+Shift+0
Windows/Linux: Ctrl+Shift+0

Step 3: Start the Proxy Server

Install and run the proxy:

# Or, use your favourite Python env manager
conda create -n dcp python=3.10 -y
conda activate dcp

# Install
git clone https://github.com/yxlao/deepseek-cursor-proxy.git
cd deepseek-cursor-proxy
pip install -e .

# Run
deepseek-cursor-proxy

The proxy creates ~/.deepseek-cursor-proxy/config.yaml on first run.

This will also print the ngrok public URL. If it differs from the one in Cursor, update it in Cursor's Base URL field.

Step 4: Chat with DeepSeek in Cursor

Select deepseek-v4-pro in Cursor and use chat or agent mode as usual.

How It Works

DeepSeek's thinking mode requires reasoning_content from assistant messages in tool-call sequences to be passed back in later requests. Cursor may omit this field, causing DeepSeek to return a 400 error. This proxy sits between Cursor and DeepSeek (Cursor → ngrok → proxy → DeepSeek API) and repairs requests when it has the exact original reasoning cached.

Core fix: every DeepSeek response, streaming or non-streaming, has its reasoning_content stored in a local SQLite cache keyed by message signature, tool-call ID, and tool-call function signature. On outgoing thinking-mode requests, the proxy restores missing reasoning_content for tool-call-related assistant messages and sends the complete history to DeepSeek. If the cache is cold, such as after a proxy restart, it returns a local error instead of fabricating reasoning.
Multi-conversation isolation: cache keys are scoped by a SHA-256 hash of the canonical conversation prefix (roles, content, tool calls, excluding reasoning_content) plus the upstream model/configuration and an API-key hash. Concurrent or interleaved threads with different histories get different scopes, so reused tool-call IDs do not collide. Byte-identical cloned histories are indistinguishable unless Cursor sends a differentiating history.
DeepSeek prefix caching compatibility: the proxy does not inject synthetic thread IDs, timestamps, or cache-control messages into the prompt. When it restores cached reasoning, it restores the exact original string, preserving repeated prefixes for DeepSeek's automatic best-effort context cache.
Additional compatibility fixes: the proxy converts legacy functions/function_call fields to tools/tool_choice, preserves required and named tool-choice semantics, normalizes reasoning_effort aliases per DeepSeek docs, strips mirrored <think> blocks from assistant content, converts multi-part content arrays to plain text, logs DeepSeek prompt-cache usage when available, and mirrors reasoning_content into Cursor-visible <think>...</think> blocks for thinking display.

Debugging

Run with verbose output:

deepseek-cursor-proxy --verbose

Run without ngrok for local curl testing:

PROXY_NGROK=false deepseek-cursor-proxy --port 9000 --verbose

Use another config file:

deepseek-cursor-proxy --config ./dev.config.yaml

Clear the local reasoning cache:

deepseek-cursor-proxy --clear-reasoning-cache

Run tests:

PYTHONPATH=src python -m unittest discover -s tests

6.1 KiB Raw Blame History