Using DeepSeek Models in Cursor: Complete Setup Guide

DeepSeek models offer powerful reasoning capabilities at competitive prices. Cursor supports DeepSeek integration through multiple methods, from native API support to local deployment via Ollama. This guide covers all approaches to get DeepSeek working in your Cursor workflow.

Available DeepSeek Models

Model	Context	Best For	Cursor Support
DeepSeek V3	64K	General coding, chat	Native
DeepSeek R1	64K	Reasoning, math, logic	Native
DeepSeek V4	128K	Complex analysis	Via Ollama/Proxy
DeepSeek Coder	16K	Code-specific tasks	Via OpenRouter

Method 1: Native DeepSeek Support (Recommended)

Cursor has built-in support for DeepSeek V3 and R1.

Step 1: Get API Key

1. Visit DeepSeek Platform
2. Create an account or sign in
3. Navigate to "API Keys"
4. Generate a new key and copy it

Step 2: Configure in Cursor

1. Open Cursor Settings (Cmd/Ctrl + ,)
2. Go to "Models" or "AI Features"
3. Find "Custom API Key" or "Add Provider"
4. Select "DeepSeek" as the provider
5. Paste your API key
6. Save settings

Step 3: Select DeepSeek Model

In any chat or Composer session:

1. Click the model selector (top of chat)
2. Choose "DeepSeek V3" or "DeepSeek R1"
3. Start coding

Method 2: Using DeepSeek via OpenRouter

For models not natively supported, use OpenRouter as a bridge.

Setup

1. Get an OpenRouter API key
2. In Cursor Settings, add OpenRouter as a provider:

{
  "customModels": [
    {
      "name": "deepseek/deepseek-chat",
      "provider": "openrouter",
      "apiKey": "your-openrouter-key"
    }
  ]
}

Available OpenRouter DeepSeek Models

deepseek/deepseek-chat          # DeepSeek V3
deepseek/deepseek-reasoner     # DeepSeek R1
deepseek/deepseek-coder        # DeepSeek Coder

Method 3: Local DeepSeek with Ollama

Run DeepSeek locally for privacy and offline use.

Install Ollama

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows
# Download from https://ollama.com/download

Pull DeepSeek Model

# DeepSeek Coder (recommended for coding)
ollama pull deepseek-coder:6.7b

# For larger context (requires more RAM)
ollama pull deepseek-coder:33b

Configure Cursor for Ollama

1. Ensure Ollama is running:

ollama serve

2. In Cursor Settings, add local model:

{
  "customModels": [
    {
      "name": "deepseek-coder",
      "provider": "ollama",
      "baseUrl": "http://localhost:11434"
    }
  ]
}

Troubleshooting Ollama Connection

If Cursor can't connect to Ollama:

# Check Ollama is running
curl http://localhost:11434/api/tags

# Set environment variable for Ollama host
export OLLAMA_HOST=0.0.0.0

# On macOS, allow network access
launchctl setenv OLLAMA_HOST "0.0.0.0"

Method 4: DeepSeek V4 via LiteLLM Proxy

For DeepSeek V4 Pro (cloud), use LiteLLM as a proxy.

Setup LiteLLM

pip install litellm

Create litellm_config.yaml:

model_list:
  - model_name: deepseek-v4
    litellm_params:
      model: deepseek/deepseek-chat-v4
      api_key: os.environ/DEEPSEEK_API_KEY

Run the proxy:

litellm --config litellm_config.yaml --port 8000

Configure Cursor

{
  "customModels": [
    {
      "name": "deepseek-v4",
      "provider": "openai-compatible",
      "apiKey": "sk-any",
      "baseUrl": "http://localhost:8000"
    }
  ]
}

Handling Reasoning Content

DeepSeek R1 outputs reasoning content that needs special handling.

The Problem

R1 returns both reasoning and final answer:

{
  "choices": [{
    "message": {
      "content": "Final answer here...",
      "reasoning_content": "Let me think... Step 1... Step 2..."
    }
  }]
}

Solution: Filter with Proxy

Use a simple proxy to strip reasoning content:

# deepseek_proxy.py
from flask import Flask, request, Response
import requests

app = Flask(__name__)

@app.route('/v1/chat/completions', methods=['POST'])
def proxy():
    resp = requests.post(
        'https://api.deepseek.com/v1/chat/completions',
        headers={'Authorization': f'Bearer {API_KEY}'},
        json=request.get_json()
    )
    data = resp.json()
    # Strip reasoning_content
    for choice in data.get('choices', []):
        msg = choice.get('message', {})
        msg.pop('reasoning_content', None)
    return Response(
        json.dumps(data),
        mimetype='application/json'
    )

Optimizing DeepSeek for Coding

System Prompt

Add this to your Cursor rules for better DeepSeek performance:

When using DeepSeek models:
- Be explicit about file paths and line numbers
- Request step-by-step reasoning for complex tasks
- Use structured output formats (JSON, markdown tables)
- Specify expected response length

Temperature Settings

Task	Recommended Temperature
Code generation	0.1 - 0.3
Debugging	0.2 - 0.4
Creative exploration	0.5 - 0.7
Code review	0.1 - 0.2

Cost Comparison

Model	Input/1M tokens	Output/1M tokens
DeepSeek V3	$0.14	$0.28
DeepSeek R1	$0.55	$2.19
GPT-4o	$5.00	$15.00
Claude 3.5 Sonnet	$3.00	$15.00

Prices as of 2026. DeepSeek is significantly more cost-effective.

Troubleshooting

Issue	Solution
"Model not available"	Check API key is valid and has credits
Slow responses	Use smaller model or enable streaming
Reasoning content visible	Use proxy or filter (see above)
Ollama connection refused	Ensure Ollama is running and port is accessible
Chinese output	Add "Respond in English" to your prompt

Quick Reference

# Test DeepSeek API
curl https://api.deepseek.com/v1/chat/completions \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-chat",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

# List Ollama models
ollama list

# Check Ollama status
curl http://localhost:11434/api/tags

Related Resources

• How to Use Custom Models in Cursor
• Cursor Model Selection Guide
• Cursor Local LLM Setup