SkillAgentSearch skills...

deepseek-search-everywhere-mcp

MCP server for web search, page fetch to Markdown, structured extraction and cited deep research via the DeepSeek Search & Fetch API. Works with Claude Code, Cursor, Cline, Windsurf, Zed and OpenCode.

Install / Use

claude mcp add radoxtech -- npx -y github:radoxtech/deepseek-search-everywhere-mcp

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

74/100

Supported Platforms

Claude Code
Claude Desktop
Cursor
Windsurf
Cline
Zed

Tags

Our assessment of deepseek-search-everywhere-mcp

deepseek-search-everywhere-mcp scores 74/100 on our quality scale, 839th of 944 AI & Machine Learning skills we index.

Its MCP Server is 43 KB long, well organised into 41 sections with 14 code examples: long enough that it reads more like full documentation than a focused instruction file, which agents can find harder to follow.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
21/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so deepseek-search-everywhere-mcp is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

deepseek-search-everywhere-mcp compared with similar skills

All 4 of these similar skills score higher than deepseek-search-everywhere-mcp; compare them before choosing.

SkillScoreStarsUpdatedFormat
deepseek-search-everywhere-mcp (this skill)by radoxtech743todayMCP Server
claude-memby thedotmack100118.5ktodayCLAUDE.md
Agent-Reachby Panniantong100113.3k12d agoCLAUDE.md
Understand-Anythingby Egonex-AI10093.3k1d agoCLAUDE.md
hyperframesby heygen-com10087.1ktodayCLAUDE.md

Frequently asked questions

How do I install deepseek-search-everywhere-mcp?
Run claude mcp add radoxtech -- npx -y github:radoxtech/deepseek-search-everywhere-mcp. The install tabs above show the steps for each supported agent.
Which AI agents does deepseek-search-everywhere-mcp work with?
It is written for Claude Code, Claude Desktop, Cursor, Windsurf, Cline and Zed, as a MCP Server file. Other agents that read the same format can often use it too.
Is deepseek-search-everywhere-mcp safe to use?
It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is deepseek-search-everywhere-mcp still maintained?
The repository was last updated today, so deepseek-search-everywhere-mcp is actively maintained.
<p align="center"> <img src="assets/banner.jpg" alt="DeepSeek Search Everywhere MCP banner" width="100%"> </p>

DeepSeek Search Everywhere MCP

<p align="center"> <a href="https://pypi.org/project/deepseek-search-everywhere-mcp/"><img src="https://img.shields.io/pypi/v/deepseek-search-everywhere-mcp.svg" alt="PyPI version"></a> <a href="https://pypistats.org/packages/deepseek-search-everywhere-mcp"><img src="https://img.shields.io/pypi/dm/deepseek-search-everywhere-mcp?color=blue&label=pypi%20downloads" alt="PyPI downloads per month"></a> <a href="https://pypi.org/project/deepseek-search-everywhere-mcp/"><img src="https://img.shields.io/pypi/pyversions/deepseek-search-everywhere-mcp" alt="Supported Python versions"></a> <a href="https://registry.modelcontextprotocol.io"><img src="https://img.shields.io/badge/MCP%20Registry-listed-brightgreen.svg" alt="Listed in the official MCP Registry"></a> <a href="https://github.com/radoxtech/deepseek-search-everywhere-mcp/actions/workflows/ci.yml"><img src="https://github.com/radoxtech/deepseek-search-everywhere-mcp/actions/workflows/ci.yml/badge.svg" alt="CI status"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT"></a> <a href="https://glama.ai/mcp/servers/radoxtech/deepseek-search-everywhere-mcp"><img src="https://glama.ai/mcp/servers/radoxtech/deepseek-search-everywhere-mcp/badges/score.svg" alt="Glama score"></a> <a href="https://m8ven.ai/mcp/radoxtech-deepseek-search-everywhere-mcp-z1tjri?s=readme"><img src="https://m8ven.ai/badge/mcp/radoxtech-deepseek-search-everywhere-mcp-z1tjri?v=015f81e910190f6de7a794111dc8f011" alt="M8ven Score"></a> </p>

DeepSeek Search Everywhere is a Model Context Protocol (MCP) server that gives AI agents web search, page fetch (web scraping to clean Markdown), structured data extraction and deep research with numbered citations, all through the DeepSeek Search & Fetch API. It runs over stdio or authenticated HTTP and works with Claude Code, Claude Desktop, Cursor, Cline, Windsurf, Zed, OpenCode and any other MCP client.

You need one credential, a DeepSeek API key, and you pay DeepSeek directly per request: there is no separate subscription. It is an alternative to Tavily, Perplexity, Firecrawl and Exa for agents that need live web access, and about 10x cheaper than Tavily for basic search; see Comparison for the numbers.

<!-- mcp-name: io.github.radoxtech/deepseek-search-everywhere-mcp -->

Not affiliated with DeepSeek. This is an independent, community-maintained project.

Demo

<p align="center"> <img src="assets/demo.gif" alt="OpenCode terminal session calling deepseek_search, deepseek_fetch_many, deepseek_extract and deepseek_research" width="100%"> </p> <p align="center"> <em>OpenCode calling <code>deepseek_search</code>, <code>deepseek_fetch_many</code>, <code>deepseek_extract</code> and <code>deepseek_research</code>. The same tools work in every MCP client.</em> </p>

Key features

  • Web search with snippets or full page text, domain and date filters, language selection and optional query rewrite.
  • Page fetch to Markdown for one URL or up to 10 in parallel, with section, snippet, offset and find-text selectors for long pages.
  • Structured extraction: pull named fields or a JSON schema from a page; each value must be backed by a verbatim quote from the page or it is returned as null.
  • Deep research with citations: four depth tiers and academic/technical source modes produce a report with [n] citations mapped to a sources array, with number verification. The server writes the report, or your agent writes it from the ranked sources (synthesis="client") and the server verifies it. See Deep research.
  • Cost reporting: every response carries cost_usd and, where an LLM is involved, cost_breakdown; an optional daily budget blocks overspending.
  • Hardened against prompt injection, with an egress guard, static SSRF filtering and no telemetry. See Security.
  • Small footprint: pure Python, no local browser, uvx zero-install, 9 tools.

Tools

| Tool | What it does | Key parameters | |---|---|---| | deepseek_search | Search the web for one query; returns ranked results with snippets, or full page text | query, count (1-50, default 5), content (snippets or full), lang, include_domains, exclude_domains, after, before, exclude_terms, exact_phrases, rewrite, max_total_chars (default 20000), refresh | | deepseek_search_many | Run 1-5 searches concurrently; partial results are returned if the deadline hits, and unique queries past the fifth are skipped and listed in skipped_queries | queries, plus the filters of deepseek_search, max_total_chars (default 20000), deadline | | deepseek_fetch | Fetch one known URL as Markdown | url, max_chars (default 20000), one optional selector: section, query (+max_snippets), offset+limit, or find_text; strip_boilerplate, fallback_to_search, refresh | | deepseek_fetch_many | Fetch 1-10 URLs concurrently (up to 5 at a time) | urls, max_chars_per_url (default 4000), max_total_chars (default 20000), query, deadline | | deepseek_extract | Extract structured JSON from one page, with evidence quotes | url, fields or schema, instruction, max_chars (default 40000) | | deepseek_extract_many | Extract structured JSON from 1-10 pages concurrently | urls, fields or schema, instruction, max_chars_per_url, max_total_chars, deadline | | deepseek_research | Multi-angle research: decomposes the topic, searches, reads sources and writes a cited report | topic (max 500 chars), depth (low, medium, high, max; default medium), sources (any, academic, technical; default any), lang, unverified (flag or drop; default flag), synthesis (server or client; default server), strip_boilerplate, refresh | | deepseek_check_research | Continue a research job: poll it, read the sources of a synthesis="client" job page by page, or submit your report for verification | job_id, page, report | | deepseek_check_balance | Read-only: DeepSeek account balance and session metrics | none |

Details that matter when calling the tools:

  • query and each entry of queries are limited to 128 characters and accept site:, -site:, "phrase", lang:, after: and before: operators. The explicit filter parameters (include_domains, exclude_domains, exclude_terms, exact_phrases, after, before) are compiled into operators and budget-checked against that limit together with the query. Dates use YYYY-MM-DD.
  • content="snippets" (default) returns a short snippet per result; content="full" returns page text and fetches pages whose search content is missing, one extra request per fetched page, which is reflected in requests_used and cost_usd.
  • deepseek_fetch selectors are mutually exclusive: combining section, query, offset/limit and find_text returns bad_request. query returns focused excerpts around matching terms (lexical matching, no LLM call). offset/limit page through long text and report total_chars.
  • deepseek_fetch_many skips blanks, fetches duplicate URLs once, keeps input order and reports a failing URL as an item with ok: false; the top-level envelope sets partial: true.
  • deepseek_extract takes either fields (list of names) or schema (JSON schema), never both. Fields whose evidence quote cannot be found in the page are set to null with a note; extraction_truncated is raised when the model output was cut off.
  • lang is an ISO code such as en, de, pt-BR. rewrite=true translates the query to English with deepseek-flash before searching and is skipped when lang is set.
  • strip_boilerplate=true removes known site interface text (for example the localized Reddit UI and English cookie-consent dialogs at the start of a page) without touching article text or code blocks. Search results always get this treatment.
  • fallback_to_search=true (fetch tools) retrieves a document through the search index when a direct fetch returns 404 for a PDF or Reddit URL; only a result for the same page is accepted, and the result is marked "source": "search_index".

Example calls:

deepseek_search(query="rust async runtime comparison", count=5)
deepseek_search(query="gobierno de España presupuestos 2026", lang="es")
deepseek_search(query="tokio runtime", include_domains=["docs.rs"], content="full")
deepseek_fetch(url="https://docs.rs/tokio/latest/tokio/", max_chars=15000)
deepseek_fetch(url="https://example.com/long-article", query="rate limits")
deepseek_fetch_many(urls=["https://example.com/a", "https://example.com/b"])
deepseek_extract(url="https://news.ycombinator.com", fields=["top_story", "points", "author"])
deepseek_research(topic="Raft vs Paxos consensus: trade-offs in practice", depth="high")
deepseek_check_research(job_id="<job_id from a running response>")
deepseek_check_balance()

How agents should use this server

Pick the tool by what the task needs:

| Task | Tool | |---|---| | A question that needs several sources: comparison, "how does X work", state of the art, pros and cons, a decision | deepseek_research (depth low to max; sources="academic" or "technical" when it fits) | | One fact, a recent event, a page to find | deepseek_search (snippets first, content="full" only when the page text is needed) | | Several independent searches at once | deepseek_search_many (up to 5 queries) | | A known URL | deepseek_fetch; pass query= to get only the relevant excerpts | | Several known URLs | deepseek_fetch_many | | Specific fields or a JSON schema from a page | deepseek_extract or deepseek_extract_many | | Account balance and session usage | deepseek_check_balance |

Research jobs have three states, and the response says which one applies:

  • status: "done": the report is in report and its [n] citations are the ids in sources.
  • status: "running": call deepseek_check_research(job_id) until it is done; do not start the same research again.
  • status: "awaiting_report" (only with synthesis="client"): read the sources (context, further pages with page=), write the report and send it with deepseek_check_research(job_id, report=...).

Common mistakes to avoid: a query plus its filters must fit in 128 characters; the deepseek_fetch selectors (section, query, offset/limit, find_text) cannot be combined; deepseek_search_many runs at most 5 unique queries and lists the rest in skipped_queries; check partial on batch results, because ok stays true when single items fail. The error codes table says which errors are worth retrying.

Quick start

  1. Create a DeepSeek API key at platform.deepseek.com and add a small balance.
  2. Install uv (or use pip; Python 3.11 or newer is required).
  3. Add the server to your MCP client. The minimal configuration is:
{
  "mcpServers": {
    "deepseek-search-everywhere": {
      "command": "uvx",
      "args": ["deepseek-search-everywhere-mcp"],
      "env": { "DEEPSEEK_API_KEY": "sk-..." }
    }
  }
}
  1. Optional: verify the key and connectivity from a terminal. This prints your balance and exits:
export DEEPSEEK_API_KEY=sk-...
uvx deepseek-search-everywhere-mcp --check

The package is on PyPI and in the official MCP Registry as io.github.radoxtech/deepseek-search-everywhere-mcp. The console script is deepseek-search-everywhere-mcp; python -m deepseek_search_everywhere_mcp is equivalent

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars3
CategoryAI
Updated22h ago
Forks0

Languages

Python

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low