deepseek-mcp
Bridges MCP clients to the DeepSeek API for chat completions and model discovery, with an optional locked-down bridge that delegates prompts to local CLI harnesses like Claude, Codex, or opencode.
README
deepseek-mcp
An MCP server that bridges CLI harnesses to the DeepSeek API. Callers (Claude Code, Codex, opencode, or any MCP client) connect over stdio and get DeepSeek chat completions — plus the option to hand work back to a local harness session.
Quickstart
One-shot install straight from the repo (clone lives in ~/.deepseek-mcp, override with DEEPSEEK_MCP_HOME). Fresh installs prompt for your API key — it opens the key page and reads a hidden paste. Updates keep your existing .env untouched:
curl -fsSL https://raw.githubusercontent.com/Korck-lab/deepseek-mcp/main/scripts/install.sh | bash
Prefer reviewing the script first, or hit terminal weirdness (no controlling tty: SSH without -t, CI)? Download then run — stdin stays yours, no pipe tricks:
curl -fsSL https://raw.githubusercontent.com/Korck-lab/deepseek-mcp/main/scripts/install.sh -o /tmp/deepseek-mcp-install.sh
bash /tmp/deepseek-mcp-install.sh
Manual clone:
git clone https://github.com/Korck-lab/deepseek-mcp
cd deepseek-mcp
npm install
npm run build
cp .env.example .env # paste your DEEPSEEK_API_KEY
npm run install:cli # interactive: pick claude/codex/opencode + scope
Then run claude mcp list, codex mcp list, or opencode mcp list to confirm the deepseek server shows Connected, and use it from that client. See Interactive installer for flags and scope details.
Features
- Chat completions — OpenAI-style
messages→ DeepSeek. Supportsdeepseek-v4-flash/deepseek-v4-proreasoning models, includingreasoning_content. - Model discovery —
list_modelsfetches available model ids from the API. - Harness bridge —
use_harnessruns a prompt in a headless host CLI (claude -p,codex exec,opencode run, or any command you configure), letting the DeepSeek model drive local agents, skills, plugins, and MCP tools configured on that host. - Locked by default — the bridge is off until you explicitly allowlist hosts.
- Zero shell — harness commands run as
spawnargument arrays, never through a shell. Prompts are positional args, so no injection surface. dotenvconfig — keys and harness commands live in.env.
Tools
| Tool | Description |
|---|---|
chat |
Send a chat completion request; returns assistant text (reasoning + answer for reasoning models). |
list_models |
List models available on the DeepSeek API. |
use_harness |
Run a prompt in a configured host CLI. Host must be allowlisted. |
list_harnesses |
Show configured harness hosts and their commands. |
Setup
npm install
cp .env.example .env # add your DEEPSEEK_API_KEY
npm run build
npm test # protocol smoke test + live API + harness bridge
Run the server:
npm start
Configuration
| Variable | Default | Purpose |
|---|---|---|
DEEPSEEK_API_KEY |
— | Required. DeepSeek API key. |
DEEPSEEK_BASE_URL |
https://api.deepseek.com |
API base URL. |
DEEPSEEK_MODEL |
deepseek-chat |
Default model for chat. |
HARNESS_ALLOW_HOSTS |
(empty = locked) | Comma-separated hosts the bridge may drive. |
HARNESS_CMD_<HOST> |
— | Headless command for a host, e.g. HARNESS_CMD_CLAUDE=claude -p. |
Example bridge config (opencode is a TUI and needs a PTY wrapper on macOS):
HARNESS_ALLOW_HOSTS=claude,codex,opencode
HARNESS_CMD_CLAUDE=claude -p
HARNESS_CMD_CODEX=codex exec --skip-git-repo-check
HARNESS_CMD_OPENCODE=script -q /dev/null opencode run
MCP client config
{
"mcpServers": {
"deepseek": {
"command": "node",
"args": ["/absolute/path/to/deepseek-mcp/dist/index.js"]
}
}
}
Interactive installer
Register the server with Claude Code, Codex, and/or opencode — no manual config editing:
npm run install:cli
Prompts ask which CLI(s) to install and the scope:
- global — your user config (
~/.claude.json,~/.codex/config.toml,~/.config/opencode/opencode.json) - project — local to this repo (Claude
--scope local,.codex/config.toml,./opencode.json; gitignored)
Optionally embeds DEEPSEEK_API_KEY into the client config (recommended for global scope, so the server works outside this repo). Scriptable:
npm run install:cli -- --clis claude,codex,opencode --scope global --embed key --yes
Flags: --clis a,b,c, --scope global|project, --embed key|skip, --yes.
Auto-versioning
Local git hooks bump the version and tag releases automatically from conventional commit messages — no manual version edits, no CI needed.
npm run hooks:install # git config core.hooksPath .githooks (per-repo)
On every commit it reads the commit message and:
| Commit type | Example | Bump |
|---|---|---|
| breaking | feat!: drop node 18, or BREAKING CHANGE in body |
major |
feat: |
feat: add installer |
minor |
fix: / perf: |
fix: opencode pty wrapper |
patch |
| anything else | docs:, chore:, test:, refactor: |
none |
After the commit is created, a post-commit hook reads the real message from COMMIT_EDITMSG (the message is not available earlier — pre-commit runs before it exists), bumps package.json (+package-lock.json), amends the commit so the bump is included, then creates an annotated tag vX.Y.Z. Merge commits never re-bump; amends do re-bump (tag already exists, so it just skips). Run npm version manually at any time to override.
Security
- The bridge is disabled until you allowlist hosts. With
HARNESS_ALLOW_HOSTSempty,use_harnesserrors. - Harness commands run without a shell; the prompt is a single positional argument.
- The API key lives only in
.env(gitignored) and is sent only to the DeepSeek API. - Each harness call spawns a fresh headless session — no access to the calling client's session state. Note that a harness CLI runs with its own configured tools and credentials; only allowlist hosts you trust.
Development
npm run build— compilesrc/→dist/npm run dev— run from source withtsxnpm test— MCP protocol smoke test (handshake, tools/list, live chat, model list, harness bridge)
Requires Node.js >= 20.
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.