Deepseek Thinker MCP Server for Claude Desktop

Overview

This MCP (Model Context Protocol) server acts as a bridge between MCP-enabled AI clients (for example, Claude Desktop) and Deepseek reasoning services or a local Ollama model. Instead of calling Deepseek or Ollama directly from the client, the client speaks the MCP protocol to this local server. The server then forwards requests to your chosen backend (Deepseek cloud API or a locally hosted Ollama instance), translates responses, and exposes MCP-compatible tool actions back to the client.

This architecture is useful when you want to:

• Integrate Deepseek reasoning into MCP-capable UIs without modifying the client.
• Use a local Ollama model for private or offline inference while preserving the MCP workflow.
• Centralize configuration and credentials (API keys, model selection) in one place so multiple clients can use the same reasoning backend.

Features

• MCP-compatible HTTP server for seamless integration with Claude Desktop and other MCP clients
• Proxying to Deepseek API using an API key and configurable base URL
• Optional local Ollama integration for on-premise/private inference
• Configurable host/port and backend selection via environment variables or CLI
• Simple deployment: Node/npm workflow and Docker-friendly build/run pattern
• Health and discovery endpoints for client registration and monitoring

Installation / Configuration

Clone and run the server locally, or build a container.

Basic local install (Node.js):

git clone https://github.com/ruixingshi/deepseek-thinker-mcp.git
cd deepseek-thinker-mcp
npm install
npm run build
# run with environment variables (examples below)
npm start

Build and run with Docker:

# build image
docker build -t deepseek-thinker-mcp .

# run container (example using Deepseek)
docker run -p 3456:3456 \
  -e BACKEND=deepseek \
  -e DEEPSEEK_API_KEY=your_api_key_here \
  -e DEEPSEEK_BASE_URL=https://api.deepseek.example \
  deepseek-thinker-mcp

Configure via environment variables (examples):

# Common settings
PORT=3456
HOST=127.0.0.1
BACKEND=deepseek  # or "ollama"

# Deepseek backend
DEEPSEEK_API_KEY=sk-...
DEEPSEEK_BASE_URL=https://api.deepseek.example

# Ollama backend (optional)
OLLAMA_URL=http://127.0.0.1:11434
OLLAMA_MODEL=your-model-name

Environment variable reference

Point your MCP-capable client (e.g., Claude Desktop) at the server URL:

• Example MCP URL: http://127.0.0.1:3456/

Check the server health endpoint (if exposed):

curl http://127.0.0.1:3456/health

Available Resources

The server exposes MCP discovery and runtime endpoints so clients can register tools and execute actions. Commonly provided resources include:

• MCP discovery endpoint (HTTP) — the client discovers tools and metadata via the server’s MCP discovery route.
• Tool actions — Deepseek-backed reasoning tools (search, structured reasoning) appear to the client as executable actions.
• Health and metrics endpoints for local monitoring.

Note: Exact path names and the shape of the MCP metadata are MCP protocol details; the server implements MCP-compatible discovery so MCP clients can register and use available tools without additional changes.

Use Cases

1. Add Deepseek reasoning to Claude Desktop

   • Run the MCP server locally, set BACKEND=deepseek and DEEPSEEK_API_KEY.
   • Point Claude Desktop’s MCP URL to http://127.0.0.1:3456/.
   • Claude Desktop will list the Deepseek tools as MCP tools and can call them directly for advanced reasoning flows (search, multi-step reasoning, tool chaining).

2. Private/local model inference with Ollama

   • Set BACKEND=ollama and configure OLLAMA_URL/OLLAMA_MODEL.
   • Use the same local MCP URL for Claude Desktop; requests are routed to your local Ollama model for inference—useful for private data or offline development.

3. Developer workflows: code reasoning and triage

   • Integrate Deepseek-powered reasoning into a developer workstation where Claude Desktop acts as an assistant.
   • Use the reasoning tools to analyze code diffs, propose refactors, or summarize repo history by combining MCP tool calls with client UIs.

4. Research and summarization pipelines

   • Use Deepseek for structured search and long-form reasoning while invoking it through the MCP server from your preferred client.
   • Combine Deepseek results with local post-processing (via Ollama or other tools) to build reproducible research workflows.

Tips for Developers

• Keep API keys and secrets out of version control; use environment variables or a secrets manager.
• If you run multiple MCP servers, use unique ports and host entries and register each server separately in the client.