DeepSeek MCP Server — Claude Code MCP server: what it does
Wires DeepSeek models into your coding agent: chat with images, FIM code completion, model discovery and balance checks.
claude mcp add deepseek -e DEEPSEEK_API_KEY=<DEEPSEEK_API_KEY> -- npx -y deepseek-mcp-serverWhat DeepSeek MCP Server does
The server hands your agent eleven tools. chat_completion covers text or visual chat with thinking controls, function tools and memory keyed by conversation_id. create_response makes stateless Responses API calls, and completion handles FIM through /beta/completions. Alongside those sit list_models, get_user_balance and four file operations: upload, list, retrieve, delete.
You need Node.js 20 or newer and a DeepSeek key in DEEPSEEK_API_KEY. Thinking is on by default, but forced tool choice in chat only works with thinking disabled — worth checking if your function calls quietly stop firing. FIM is beta and non-thinking, and it carries its own output limit, so the chat ceiling does not apply there.
create_response keeps no history; send the full input each time. web_search runs server-side only with deepseek-v4-pro, while deepseek-flash ignores it. There are no standalone Search or Fetch tools here, and MCP_TOOL_RATE_LIMIT caps MCP invocations rather than what you spend at the provider.
What it helps with
- call a DeepSeek model from the agent without leaving the editor
- fill in code from the surrounding context with FIM
- upload an image once and reuse its
file_id - see which models the account can reach right now
- check the remaining balance before a long session
Example prompts for the agent
- List the DeepSeek models available to my account with their context limits
- Ask deepseek-v4-pro what is wrong with this file and return the full answer
- Upload this PNG and remember the file_id for later calls
How to install DeepSeek MCP Server
Platform CLI
claude mcp add deepseek -e DEEPSEEK_API_KEY=<DEEPSEEK_API_KEY> -- npx -y deepseek-mcp-serverReplace <…> with your own values
Platform CLI
codex mcp add deepseek-mcp-server --env DEEPSEEK_API_KEY=<DEEPSEEK_API_KEY> -- npx -y deepseek-mcp-serverReplace <…> with your own values
Config file
{
"mcpServers": {
"deepseek-mcp-server": {
"command": "npx",
"args": [
"-y",
"deepseek-mcp-server"
],
"env": {
"DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"deepseek-mcp-server": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"deepseek-mcp-server"
],
"env": {
"DEEPSEEK_API_KEY": "<DEEPSEEK_API_KEY>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e DEEPSEEK_API_KEY=<DEEPSEEK_API_KEY> deepseek-mcp-server npx -y deepseek-mcp-serverReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about DeepSeek MCP Server
How do I install DeepSeek MCP Server in Claude?
Run claude mcp add deepseek -e DEEPSEEK_API_KEY=<DEEPSEEK_API_KEY> -- npx -y deepseek-mcp-server. Use your own key; Node.js 20 or newer has to be installed first.
What does DeepSeek MCP Server do?
It gives your agent access to the DeepSeek API: text and image chat, the Responses API, FIM completion, file handling, model listing and account balance.
How is this different from calling the DeepSeek API directly?
It wraps the calls as MCP tools with output schemas and annotations, and only returns full provider payloads when include_raw_response=true. It also keeps bounded in-memory conversations.
