deepseek-search
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@deepseek-search search the web for the latest Node.js LTS release notes"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
deepseek-search-fetch-mcp
DeepSeek 官方 Search / Fetch API 的 stdio MCP 服务器。标准 MCP 协议,Claude Code、Codex、DeepSeek Harness、WorkBuddy、Cursor 都能直接挂载。
把这段丢给你的 Agent
帮我安装一个 MCP 服务器,名称 deepseek-search。
- 来源:npm 公开包 deepseek-search-fetch-mcp,stdio 传输
- 启动命令:npx -y deepseek-search-fetch-mcp
Windows 上若客户端不经过 shell,用:cmd /c npx -y deepseek-search-fetch-mcp
- 环境变量:DEEPSEEK_API_KEY=<我的 Key>(我没有的话先问我)
- 请写进你所在客户端自己的 MCP 配置并重载:
Claude Code 用 claude mcp add;Codex 用 codex mcp add 或 ~/.codex/config.toml;
DeepSeek Harness 用官方 @deepseek-ai/dsh-mcp-client + cordis.patch.yml;
WorkBuddy 用 ~/.workbuddy/mcp.json;Cursor 用 .cursor/mcp.json
- 装好后确认 deepseek_search 和 deepseek_fetch 两个工具可调用;
启动失败就把 stderr 原文贴给我Related MCP server: deepseek-web
客户端
客户端 | 配置位置 | 备注 |
Claude Code |
| Windows 命令前加 |
Codex |
| CLI、IDE、桌面端共用一份 |
DeepSeek Harness |
| 官方 |
WorkBuddy |
| 保存后要在连接器里点「信任」 |
Cursor |
| 标准 |
Claude Code
# Windows
claude mcp add --scope user --env DEEPSEEK_API_KEY=sk-你的Key deepseek-search -- cmd /c npx -y deepseek-search-fetch-mcp# macOS / Linux
claude mcp add --scope user --env DEEPSEEK_API_KEY=sk-你的Key deepseek-search -- npx -y deepseek-search-fetch-mcpCodex
# Windows
codex mcp add deepseek-search --env DEEPSEEK_API_KEY=sk-你的Key -- cmd /c npx -y deepseek-search-fetch-mcp# macOS / Linux
codex mcp add deepseek-search --env DEEPSEEK_API_KEY=sk-你的Key -- npx -y deepseek-search-fetch-mcpDeepSeek Harness
先按宿主版本装官方 MCP 客户端(npm 的 latest 停在旧版,必须 pin):
dsh --version
dsh plugin --profile web add @deepseek-ai/dsh-mcp-client@<上一步输出的版本>再写入 ~/.dsh/profiles/web/cordis.patch.yml(或 ~/.dsh/cordis.patch.yml):
- insert:
- id: mcp-deepseek-search
name: '@deepseek-ai/dsh-mcp-client'
config:
serverName: deepseek-search
transport: stdio
command: npx
args: ['-y', 'deepseek-search-fetch-mcp']
env:
DEEPSEEK_API_KEY: !!js process.env.DEEPSEEK_API_KEY重启 DSH 后工具名是 mcp__deepseek-search__deepseek_search。DSH 默认工具超时 60s,本服务器 Fetch 最长 55s,不会被宿主先掐断。
WorkBuddy
{
"mcpServers": {
"deepseek-search": {
"command": "npx",
"args": ["-y", "deepseek-search-fetch-mcp"],
"env": { "DEEPSEEK_API_KEY": "sk-..." }
}
}
}Windows 拉不起 npx 时,改成 "command": "cmd", "args": ["/d", "/s", "/c", "npx -y deepseek-search-fetch-mcp"]。
Cursor
{
"mcpServers": {
"deepseek-search": {
"command": "npx",
"args": ["-y", "deepseek-search-fetch-mcp"],
"env": { "DEEPSEEK_API_KEY": "sk-..." }
}
}
}工具
工具 | 参数 | 返回 |
|
| 标题、URL、摘要、发布日期、权威度、相关性 |
|
| 标题 + Markdown 正文 |
为什么是它
一手证据:返回结构化搜索结果,不是模型二次加工的答案;按次计费,不烧模型 token。
便宜:5 元 / 1000 次成功请求,失败、空结果、超时不收费。
不闯祸:本地 4 次/秒限速、429 退避、45s/55s deadline、内网地址拦截;Key 只从环境变量或指定文件读取。
开发
npm test · npm run verify:package · npm run smoke:list · MIT · English
Available Tools
2 toolsdeepseek_fetchDeepSeek Fetch PageARead-only
Fetch a single public HTTP/HTTPS page with DeepSeek and return its title and Markdown content. Use for a specific URL when search summaries are insufficient.
Not a crawler: for whole-site crawl/map or JS-heavy pages, use a dedicated crawler.
Localhost/private/internal addresses and URLs with credentials are rejected.
Content is truncated at max_chars.
Inputs: url (required), max_chars.
deepseek_searchDeepSeek Web SearchARead-only
Search the live web with DeepSeek's search engine. Returns ranked results with title, URL, snippet, long summary, publish date, site name, authority (0-4; >=3 highly authoritative), and relevance score (0-3; >=1.5 high quality).
Query limits and operators:
Query is limited to 128 Unicode characters; longer queries are truncated with a warning.
site:example.com, -site:example.com, -word, "exact phrase", -"phrase", lang:zh,en, after:YYYY-MM-DD, before:YYYY-MM-DD.
Guidance:
Best for Chinese-language and China-focused queries; pair it with another provider for English technical documentation.
need_content=false by default; use deepseek_fetch on selected URLs when full page text is needed.
Empty results are a normal successful search, not an error.
Inputs: query (required), count, freshness, need_content, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
The two tools have crisply distinct roles: one searches for ranked results by query, the other retrieves full Markdown content for a single URL. The search description explicitly points to the fetch tool when summaries are insufficient, so an agent will never misselect between them.
Both names follow the same deepseek_<verb> pattern (deepseek_search, deepseek_fetch), using consistent snake_case with an unambiguous action verb. Zero variation in convention across the set.
Two tools is slightly under the typical 3-15 band, but search+fetch is the canonical minimal-yet-complete pair for a web retrieval server, and each tool clearly earns its place. Nothing feels padded or missing at the count level.
Query-based search and single-URL fetching together cover the core retrieve-and-read lifecycle for this domain, and out-of-scope operations (whole-site crawl/map, JS-heavy rendering) are explicitly delegated to dedicated crawlers rather than left as silent gaps. Minor gap: no pagination or multi-URL batch fetch, though agents can loop over fetch calls.
Maintenance
Related MCP Connectors
Scrape, crawl and search the web for AI agents via MCP.
Live AI-native web search with citations. One tool for every MCP client. Flat per-request pricing.
Free web search for AI agents. No API key required. Hosted MCP in active development.
Web search API and MCP server for developers building AI agents.
Related MCP Servers
- AlicenseAqualityCmaintenanceEnables MCP-compatible clients to search the web and fetch HTTP/HTTPS pages as Markdown, returning result metadata such as titles, URLs, dates, relevance, and summaries.219 npm1MIT
- AlicenseAqualityCmaintenanceEnables MCP clients to perform web searches and fetch HTTP(S) pages as Markdown through Search and Fetch APIs.263 npmMIT
- AlicenseAqualityCmaintenanceEnables any MCP-compatible AI agent to search the live web and fetch pages as clean Markdown, including batch fetching with rate limiting, retries, and context-safe truncation.3MIT
- AlicenseAqualityBmaintenanceEnables MCP clients to run web searches and fetch pages, returning structured search results and Markdown page content through two tools.2MIT