deepseek-mcp

Overview

deepseek-mcp is a minimal stdio-based MCP server with two primary tools: 'deepseek' for fast, cheap, non-thinking tasks and 'advise' for deep reasoning. It's designed for bounded tasks where another model can reduce mechanical load, such as classification, summarization, template population, and producing reviewable candidate outputs. The server communicates via JSON-RPC, streams responses, and includes metadata like model, latency, tokens, and cost.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• deepseekdeepseek(prompt, system?)

  Fast, cheap, non-thinking processing for classification, extraction, formatting, and mechanical edits

  Try:Classify these files into doc / code / config: README.md, pyproject.toml, src/deepseek_mcp/server.pyTry:Turn this rough note into a CSV tableTry:Extract all TODOs and group them by owner
• adviseadvise(prompt, system?, effort?)

  Deep reasoning with configurable effort for architecture, tradeoffs, and second opinions

  Try:Should we build or buy analytics?Try:Advise on the architecture for our new microserviceTry:Review these design decisions for potential tradeoffs

Comparable tools

Installation

deepseek-mcp-server

pip install deepseek-mcp

{
  "mcpServers": {
    "deepseek": {
      "command": "deepseek-mcp-server",
      "args": [],
      "env": {
        "DEEPSEEK_API_KEY": "${DEEPSEEK_API_KEY}"
      }
    }
  }
}

export DEEPSEEK_API_KEY="sk-..."

FAQ

What models are used by the deepseek and advise tools?

The 'deepseek' tool uses deepseek-v4-flash for fast, non-thinking tasks, while 'advise' uses deepseek-v4-pro for deep reasoning with adjustable effort levels.

How can I test if my setup is working?

Use the CLI smoke test: `python -m deepseek_mcp check` or run a one-shot call with `python -m deepseek_mcp run "your prompt"`. You can also test directly with JSON-RPC as shown in the README.

Compare deepseek-mcp with