# arikusi/deepseek-mcp-server [Health: Active]

**Category:** 🔗 Aggregators  
**Repository:** https://github.com/arikusi/deepseek-mcp-server  
**GitHub Stars:** 26  
**npm Downloads (last month):** 2086  
**Views:** 13  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/arikusi-deepseek-mcp-server

## Description
MCP server for DeepSeek AI with chat, reasoning, multi-turn sessions, function calling, thinking mode, and cost tracking.

## Tools
Capabilities this server exposes over MCP:

- **deepseek_chat** — Chat with DeepSeek AI models with automatic cost tracking and function calling support.
- **deepseek_fim** — Fill-in-the-Middle completion. You give a `prompt` (the prefix) and an optional `suffix`, and the model completes the text in between. It is built for code completion and content infilling rather than conversation. FIM runs on DeepSeek's Beta endpoint in non-thinking mode, and the API caps output a…
- **deepseek_sessions** — Manage conversation sessions.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

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

**Requires environment variables:** `DEEPSEEK_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What arikusi/deepseek-mcp-server does

arikusi/deepseek-mcp-server connects MCP-compatible clients to DeepSeek's API. Its primary chat interface supports the current `deepseek-flash` model and `deepseek-v4-pro`, with both models offering a 1M-token context window and optional thinking mode. Older model names such as `deepseek-chat`, `deepseek-reasoner`, and `deepseek-v4-flash` remain accepted and are mapped to the current flash model behavior.

The server is aimed at conversational AI, reasoning tasks, structured generation, and coding workflows. It tracks request costs while accounting for cache hits and misses and the peak or off-peak rate that applied to each request.

## How it works

MCP clients call the server's tools rather than contacting DeepSeek directly. The `deepseek_chat` tool accepts conversation messages and options such as model selection, token limits, temperature, streaming, thinking, JSON mode, response schemas, and function-calling definitions. Thinking can be enabled or disabled per request, and reasoning effort can be selected while thinking is active.

Conversation state can be retained with a `session_id`, allowing multi-turn interactions across requests. Model fallback and circuit-breaker behavior help manage model or service failures. Three MCP resources are also exposed: `deepseek://models`, `deepseek://config`, and `deepseek://usage`.

## Setup and configuration

A DeepSeek API key and Node.js 22 or newer are required for a local installation. Install the published package globally with npm, or clone the repository, install dependencies, and build the TypeScript project. Configure the local stdio server with the `DEEPSEEK_API_KEY` environment variable through the MCP client's server settings.

A hosted endpoint is available at `https://deepseek-mcp.tahirl.com/mcp`. Clients that support HTTP transport can use it without installing Node.js; requests still use your own DeepSeek API key through an Authorization header. The server can also be self-hosted with Streamable HTTP by setting its transport to HTTP. Docker deployment is supported through the included container configuration and health checks.

## Tools and capabilities

- `deepseek_chat`: Sends chat messages to DeepSeek, with function calling for up to 128 tool definitions, streaming, thinking mode, JSON output, schema validation, and automatic cost tracking.
- `deepseek_fim`: Generates text between a supplied prefix and optional suffix. It targets code completion and content infilling rather than conversational exchanges and uses DeepSeek's Beta endpoint in non-thinking mode.
- `deepseek_sessions`: Lists, deletes, and clears stored conversation sessions.

The server includes 12 prompt templates covering tasks such as debugging, code review, and function calling. Multimodal content parts for text and images are available when multimodal support is enabled with `ENABLE_MULTIMODAL=true`.

_Full upstream README: https://allmcps.com/mcp/arikusi-deepseek-mcp-server/readme_