DeepseekMCP
If you are the rightful owner of DeepseekMCP and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.
A production-grade MCP server integrating with DeepSeek's API, featuring advanced code review capabilities, efficient file management, and API account management.
DeepSeek MCP Server
A production-grade MCP server integrating with DeepSeek's API, featuring advanced code review capabilities, efficient file management, and API account management.
Features
- Multi-Model Support: Choose from various DeepSeek models including DeepSeek Chat and DeepSeek Coder
- Dual Transport Support: Both stdio and HTTP transport modes for flexible integration
- Code Review Focus: Built-in system prompt for detailed code analysis with markdown output
- Automatic File Handling: Built-in file management with direct path integration
- API Account Management: Check balance and estimate token usage
- JSON Mode Support: Request structured JSON responses for easy parsing
- Advanced Error Handling: Graceful degradation with structured error logging
- Improved Retry Logic: Automatic retries with configurable exponential backoff for API calls
- Security: Configurable file type restrictions and size limits
- Performance Monitoring: Built-in metrics collection for request latency and throughput
Prerequisites
- Go 1.21+
- DeepSeek API key
- Basic understanding of MCP protocol
Installation & Quick Start
# Clone and build
git clone https://github.com/your-username/DeepseekMCP
cd DeepseekMCP
go build -o bin/deepseek-mcp
# Stdio Mode (Default - for Claude Desktop)
export DEEPSEEK_API_KEY=your_api_key
export DEEPSEEK_MODEL=deepseek-chat
./bin/deepseek-mcp
# HTTP Mode (for REST API integration)
export DEEPSEEK_API_KEY=your_api_key
export DEEPSEEK_HTTP_ENABLED=true
export DEEPSEEK_HTTP_PORT=8080
./bin/deepseek-mcp
# Or use command-line flags
./bin/deepseek-mcp --http --http-port 8080 --http-host localhost
Configuration
Claude Desktop (Stdio Mode)
{
"mcpServers": {
"deepseek": {
"command": "/Your/project/path/bin/deepseek-mcp",
"env": {
"DEEPSEEK_API_KEY": "YOUR_API_KEY",
"DEEPSEEK_MODEL": "deepseek-chat"
}
}
}
}
HTTP Mode Integration
For web applications, microservices, or API gateways:
# Start HTTP server
./bin/deepseek-mcp --http --http-port 8080
# Test with curl
curl -X POST http://localhost:8080/mcp/tools/call \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "deepseek_ask",
"arguments": {"query": "Explain Go channels"}
}
}'
Environment Variables
| Variable | Description | Default |
|---|---|---|
| API Configuration | ||
DEEPSEEK_API_KEY | DeepSeek API key | Required |
DEEPSEEK_MODEL | Model ID from available models | deepseek-chat |
DEEPSEEK_SYSTEM_PROMPT | System prompt for code review | Default code review prompt |
DEEPSEEK_SYSTEM_PROMPT_FILE | Path to file containing system prompt | Empty |
DEEPSEEK_MAX_FILE_SIZE | Max upload size (bytes) | 10485760 (10MB) |
DEEPSEEK_ALLOWED_FILE_TYPES | Comma-separated MIME types | [Common text/code types] |
DEEPSEEK_TIMEOUT | API timeout in seconds | 90 |
DEEPSEEK_MAX_RETRIES | Max API retries | 2 |
DEEPSEEK_INITIAL_BACKOFF | Initial backoff time (seconds) | 1 |
DEEPSEEK_MAX_BACKOFF | Maximum backoff time (seconds) | 10 |
DEEPSEEK_TEMPERATURE | Model temperature (0.0-1.0) | 0.4 |
| HTTP Transport Configuration | ||
DEEPSEEK_HTTP_ENABLED | Enable HTTP transport mode | false |
DEEPSEEK_HTTP_PORT | HTTP server port | 8080 |
DEEPSEEK_HTTP_HOST | HTTP server host | localhost |
DEEPSEEK_HTTP_ENDPOINT_PATH | MCP endpoint path | /mcp |
DEEPSEEK_HTTP_STATELESS | Enable stateless mode | true |
Example .env:
# API Configuration
DEEPSEEK_API_KEY=your_api_key
DEEPSEEK_MODEL=deepseek-chat
DEEPSEEK_SYSTEM_PROMPT="Your custom code review prompt here"
# Alternative: load system prompt from file
# DEEPSEEK_SYSTEM_PROMPT_FILE=/path/to/prompt.txt
DEEPSEEK_MAX_FILE_SIZE=5242880 # 5MB
DEEPSEEK_ALLOWED_FILE_TYPES=text/x-go,text/markdown
DEEPSEEK_TEMPERATURE=0.7
# HTTP Transport (optional)
DEEPSEEK_HTTP_ENABLED=true
DEEPSEEK_HTTP_PORT=8080
DEEPSEEK_HTTP_HOST=localhost
DEEPSEEK_HTTP_ENDPOINT_PATH=/mcp
DEEPSEEK_HTTP_STATELESS=true
Core API Tools
Currently, the server provides the following tools:
deepseek_ask
Used for code analysis, review, and general queries with optional file path inclusion.
{
"name": "deepseek_ask",
"arguments": {
"query": "Review this Go code for concurrency issues...",
"model": "deepseek-chat",
"systemPrompt": "Optional custom review instructions",
"file_paths": ["main.go", "config.go"],
"json_mode": false
}
}
deepseek_models
Lists all available DeepSeek models with their capabilities.
{
"name": "deepseek_models",
"arguments": {}
}
deepseek_balance
Checks your DeepSeek API account balance and availability status.
{
"name": "deepseek_balance",
"arguments": {}
}
deepseek_token_estimate
Estimates the token count for text or a file to help with quota management.
{
"name": "deepseek_token_estimate",
"arguments": {
"text": "Your text to estimate...",
"file_path": "path/to/your/file.go"
}
}
Supported Models
The following DeepSeek models are supported by default:
| Model ID | Description |
|---|---|
deepseek-chat | General-purpose chat model balancing performance and efficiency |
deepseek-coder | Specialized model for coding and technical tasks |
deepseek-reasoner | Model optimized for reasoning and problem-solving tasks |
Note: The actual available models may vary based on your API access level. The server will automatically discover and make available all models you have access to through the DeepSeek API.
Supported File Types
| Extension | MIME Type |
|---|---|
| .go | text/x-go |
| .py | text/x-python |
| .js | text/javascript |
| .md | text/markdown |
| .java | text/x-java |
| .c/.h | text/x-c |
| .cpp/.hpp | text/x-c++ |
| 25+ more | (See getMimeTypeFromPath in deepseek.go) |
Operational Notes
- Degraded Mode: Automatically enters safe mode on initialization errors
- Audit Logging: All operations logged with timestamps and metadata
- Security: File content validated by MIME type and size before processing
File Handling
The server handles files directly through the deepseek_ask tool:
- Specify local file paths in the
file_pathsarray parameter - The server automatically:
- Reads the files from the provided paths
- Determines the correct MIME type based on file extension
- Uploads the file content to the DeepSeek API
- Uses the files as context for the query
This direct file handling approach eliminates the need for separate file upload/management endpoints.
JSON Mode Support
For integrations that require structured data output, the server supports JSON mode:
- Structured Responses: Request properly formatted JSON responses from DeepSeek models
- Parser-Friendly: Ideal for CI/CD pipelines and automation systems
- Easy Integration: Simply set
json_mode: truein your request
Example with JSON mode:
{
"name": "deepseek_ask",
"arguments": {
"query": "Analyze this code and return a JSON object with: issues_found (array of strings), complexity_score (number 1-10), and recommendations (array of strings)",
"model": "deepseek-chat",
"json_mode": true,
"file_paths": ["main.go", "config.go"]
}
}
This returns a well-formed JSON response that can be parsed directly by your application.
Development
Command-line Options
The server supports these command-line options to override environment variables:
# API Configuration overrides
./bin/deepseek-mcp -deepseek-model=deepseek-coder
./bin/deepseek-mcp -deepseek-system-prompt="Your custom prompt here"
./bin/deepseek-mcp -deepseek-temperature=0.8
# HTTP Transport flags
./bin/deepseek-mcp --http --http-port 8080 --http-host localhost
./bin/deepseek-mcp --http --http-port 9000 --http-host 0.0.0.0
HTTP API Endpoints
When running in HTTP mode, the server provides these MCP endpoints:
| Endpoint | Method | Description | Status |
|---|---|---|---|
/mcp/initialize | POST | Initialize MCP session | ✅ Tested |
/mcp/tools/list | POST | List available tools | ✅ Tested |
/mcp/tools/call | POST | Execute tool calls | ✅ Tested |
/mcp/resources/list | GET | List resources | Available |
All endpoints use JSON-RPC 2.0 format and have been thoroughly tested with real DeepSeek API integration.
Running Tests
To run tests:
go test -v ./...
Running Linter
golangci-lint run
Formatting Code
gofmt -w .
License
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
- Fork the project
- Create your feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add some amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
