Skip to content
MCP server

Deepseek

DeepSeek chat, reasoning and sessions for MCP clients, with cost tracking

Install
Pricing
Open source
Platforms
CLI, API, self-hosted
Runs
Locally or Remotely
Version
2.4.0
Listing updated
2 days ago

About Deepseek

Deepseek is an open-source MCP server for ai models and llm tooling. It can run locally on your machine or as a hosted remote server.

This server puts DeepSeek models behind MCP so clients such as Claude Code, Gemini CLI, Cursor and Windsurf can call them. The README lists deepseek-flash (the V4.1 Flash default) and deepseek-v4-pro, both with a 1M-token context window and optional chain-of-thought thinking. Older names such as deepseek-chat and deepseek-reasoner are still accepted and sent as deepseek-flash.

Beyond plain chat it adds multi-turn sessions via a session_id, model fallback with a circuit breaker, JSON output with optional schema validation and bounded repair retries, and OpenAI-style function calling with up to 128 tool definitions. A deepseek_fim tool handles fill-in-the-middle completion, and deepseek_sessions lists, deletes and clears sessions. Cost tracking separates cache hits from misses and prices each request at the peak or off-peak rate that applied.

You bring your own DeepSeek API key. Run it locally over stdio with npx, or use the hosted streamable-HTTP endpoint at deepseek-mcp.tahirl.com, which takes the key as a Bearer token. It can also be self-hosted in Docker. The README reports 368 tests with about 92% line coverage.

Install

Run in a terminal

claude mcp add deepseek --env DEEPSEEK_API_KEY='<DEEPSEEK_API_KEY>' -- npx -y @arikusi/deepseek-mcp-server

Add to claude_desktop_config.json

{
  "mcpServers": {
    "deepseek": {
      "command": "npx",
      "args": [
        "-y",
        "@arikusi/deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "deepseek": {
      "command": "npx",
      "args": [
        "-y",
        "@arikusi/deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "deepseek": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@arikusi/deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "deepseek": {
      "command": "npx",
      "args": [
        "-y",
        "@arikusi/deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "deepseek": {
      "command": "npx",
      "args": [
        "-y",
        "@arikusi/deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
      }
    }
  }
}

Configuration

  • DEEPSEEK_API_KEYDeepSeek API keyRequired
  • DEEPSEEK_API_KEYDeepSeek API key (sent as Bearer token)Required

Tools this server provides (2)

Taken from the project documentation; the running server may differ.

deepseek_fim
Fill-in-the-middle completion between a prefix and suffix.
deepseek_sessions
List, delete and clear conversation sessions.

What it does

Two DeepSeek models
deepseek-flash and deepseek-v4-pro, each with a 1M-token context and optional thinking mode.
Multi-turn sessions
Context persists across calls through a session_id, with a tool to list, delete and clear sessions.
Fallback and circuit breaker
Falls back between models and guards against cascading failures.
Schema-validated JSON
A response_schema is checked against the output, with bounded repair retries.
Function calling
OpenAI-compatible tool use with up to 128 tool definitions.
Fill-in-the-middle
The deepseek_fim tool completes code or text between a prefix and suffix.
Peak-aware cost tracking
Reports cache hit and miss costs at the peak or off-peak rate in effect.

Good for

  • Call DeepSeek reasoning from inside Claude Code or Cursor
  • Run multi-turn chats that keep context across requests
  • Get JSON output validated against a schema
  • Complete code between a prefix and a suffix
  • Track what each DeepSeek request costs

Pricing

Open source

Free server; DeepSeek API usage billed separately by DeepSeek

Prices change. Check mcp.tahirl.com before you buy.

Deepseek FAQ

Do I need a DeepSeek API key?

Yes. It is required for both the local npm package and the hosted endpoint, where it is sent as a Bearer token.

Can I use it without installing anything?

Yes, through the hosted streamable-HTTP endpoint at https://deepseek-mcp.tahirl.com/mcp with your own key.

Which models does it use by default?

deepseek-flash is the default, next to deepseek-v4-pro.

Does it support images?

The README calls it multimodal ready: text and image content parts can be enabled with ENABLE_MULTIMODAL=true.

Is Deepseek free?

Yes. Deepseek is open source, so you can use and self-host it for free.

How do I add Deepseek to Claude Code?

Run this command in your terminal: claude mcp add deepseek --env DEEPSEEK_API_KEY='<DEEPSEEK_API_KEY>' -- npx -y @arikusi/deepseek-mcp-server

What are the best alternatives to Deepseek?

The closest alternatives to Deepseek are Claude Code Ultimate Guide, signafAIde, Toolkien, Lugh and HypePaper.

Alternatives to Deepseek

  1. MCP server for searching the Claude Code Ultimate Guide from any MCP client

    Runs locally
    • Open source
    • 6.5K GitHub stars
    • AI models and LLM tooling
  2. signafAIde

    MCP server

    Local MCP server to set up, verify and check coverage of signed agent action records

    Runs locally
    • 5 tools
    • Freemium
    • Security
  3. Toolkien

    MCP server

    One MCP gateway for all your MCP servers, with code mode, permissions and audit

    Not checked yet
    • 3 tools
    • Remote
    • Open source
    • AI models and LLM tooling
  4. Lugh

    MCP server

    Write, check and dry-run Lugh agent programs with policy, budgets and approvals

    Runs locally
    • 5 tools
    • Open source
    • AI models and LLM tooling
  5. HypePaper

    MCP server

    Research papers, benchmarks, citations and topic experts for AI assistants

    • 58 tools
    • Remote
    • Freemium
    • AI models and LLM tooling