DMontgomery40/deepseek-mcp-server

What it solves

This project provides a Model Context Protocol (MCP) server that allows AI clients (like Claude Code or Codex CLI) to interact with DeepSeek's V4.1 Flash API. It bridges the gap between MCP-compatible applications and DeepSeek's suite of capabilities, including text and visual chat, file management, and model discovery.

How it works

The server acts as a middleware that exposes DeepSeek's API functions as a set of eleven standardized MCP tools. It supports two transport methods: standard input/output (stdio) for local processes and a streamable HTTP endpoint for remote or browser-based clients. It also manages bounded in-memory conversations to maintain chat context.

Who it’s for

Developers and users of MCP-compatible AI agents and clients who want to integrate DeepSeek's high-context, multimodal (image understanding) models into their workflows.

Highlights

• Multimodal Support: Handles image understanding via URLs, base64 data, or the DeepSeek Files API.
• Comprehensive Toolset: Includes tools for chat completions, stateless responses, FIM (Fill-In-the-Middle) completion, and account balance checks.
• File Lifecycle Management: Full support for uploading, listing, retrieving, and deleting files for reusable image inputs.
• Flexible Transport: Supports both stdio and HTTP transports with origin allowlisting for security.
• 1M Token Context: Leverages DeepSeek Flash's large context window and high output token limits.