Deepseek
Route agent workflows through DeepSeek models with sessions, reasoning, tool calls, and cost visibility from one MCP server.
Overview
DeepSeek MCP is a Build-phase MCP server that connects agents to DeepSeek AI for chat, reasoning, sessions, function calling, and cost tracking.
What is this MCP server?
- DeepSeek chat and reasoning exposed as MCP tools with session support
- Function calling so agents can chain DeepSeek with other MCP capabilities
- Built-in cost tracking for API spend visibility during iteration
- Dual transport: npm stdio package @arikusi/deepseek-mcp-server and streamable-http endpoint
- Version 1.7.0 with required DEEPSEEK_API_KEY (Bearer on HTTP transport)
- Registry version 1.7.0 with two package entries (stdio npm and streamable-http)
- npm identifier @arikusi/deepseek-mcp-server
- Hosted HTTP transport URL https://deepseek-mcp.tahirl.com/mcp
What problem does it solve?
Switching agents to DeepSeek usually means custom API glue, no session discipline, and opaque token spend across tools.
Who is it for?
Solo builders standardizing on DeepSeek as the LLM backend inside Claude Code, Cursor, or Windsurf MCP stacks.
Skip if: Teams that only need Claude or OpenAI with no DeepSeek account, or builders avoiding external LLM API keys entirely.
What do I get? / Deliverables
After registration, your agent calls DeepSeek through MCP with sessions, tool use, and cost tracking on stdio or HTTP without bespoke client code.
- MCP-accessible DeepSeek chat and reasoning tools
- Session-aware agent runs with optional function calling
- Observable API cost data during agent sessions
Recommended MCP Servers
Journey fit
Model access and agent wiring are core build-phase work when you standardize which LLM backs your ship loop. This server lives under agent-tooling because it is the transport layer for chat, reasoning, and function calling—not app frontend code.
How it compares
LLM provider MCP bridge with cost tracking, not a prompt library skill or local model runner.
Common Questions / FAQ
Who is DeepSeek MCP for?
It is for indie developers and agent users who want DeepSeek as a first-class MCP model with sessions, tools, and spend visibility.
When should I use DeepSeek MCP?
Use it while building agent workflows when you need reasoning, multi-turn sessions, or function calling against DeepSeek instead of hand-rolled API scripts.
How do I add DeepSeek MCP to my agent?
Install @arikusi/deepseek-mcp-server via npm for stdio, or point your client at the streamable-http URL with DEEPSEEK_API_KEY set as documented in the registry.