argo DSH plugin

Argo is designed for AI agents to perform real-world searches across 120+ sources and 60+ domains, routing queries by language and domain. It outputs compact JSON evidence candidates with built-in credibility decomposition, saving tokens. It also includes deep research, evidence closure, and optional expert modes.

browser-web skills-workflows

How to install the argo DSH plugin

dsh plugin --profile web add "github:taxueseek/argo#main&path:packages/dsh-plugin

Copying does not run this command. Review the repository and version before installing the argo DSH plugin.

argo DSH plugin data source

argo DSH plugin snapshot date: Aug 16, 2026

discovered

What the argo DSH plugin can do

Unified search with automatic routing to vertical sources (stocks, movies, sports, etc.) backed by 120+ engines and 60+ domains.
Evidence closure: high-risk queries are tagged with `fetch_required`, and `--verify` performs one-click fact-checking with backfill of verified scores.
Deep research: decomposes broad questions into sub-questions, parallel multi-source collection, and gap analysis.
Dual-layer caching (memory LRU + SQLite) for hot queries (~10ms response) and isolation of login-dependent results.
10 MCP tools: search, research, evidence, disambiguation, fetch, screenshot, PDF, social, local file search, crawl.

Where the argo DSH plugin fits

An agent needs to check real-time A-share stock prices; Argo routes to Sina/Tencent quotes automatically.
A fact-checking agent searches for a claim and then verifies the top results with `--verify` to get a verified evidence score.
A multilingual agent queries in Japanese or Korean; Argo detects language and routes to language-friendly engines, avoiding Chinese-only sources.
A researcher agent uses deep research mode to gather evidence on a broad topic, with gap hints on missing evidence.

Who the argo DSH plugin is for

AI agent developers and users who need reliable, token-efficient web search capabilities.
Individuals and teams building MCP-based AI applications that require multilingual search and evidence verification.

argo DSH plugin limitations

Requires Python 3.10+ and PyYAML; some optional dependencies (curl_cffi, ddgs, etc.) are needed for advanced features like anti-crawling or offline search.
Login-dependent professional search (e.g., Zhihu, Xiaohongshu) is disabled by default and requires additional setup of ego-search subskill.
Part of the functionality is still evolving (v2.8.0); some engines may have limited availability or require API keys for best quality.

argo DSH plugin: from the repository README

Quoted from the taxueseek/argo README, the upstream source of the argo DSH plugin. Copyright remains with the original authors.

> 简单来说：前三种方案解决「**人**找信息」，Argo 解决「**Agent 及搜索核查于一身，具备一条龙的搜索服务**」。差别不在界面，在交付物，给人看的叫总结页或链接清单，给 Agent 的应是能排序、能复核、不撑爆上下文的优质内容，更可靠的搜索信息。 <p align="center"> <img src="assets/readme/why-better.svg" width="100%" alt="左侧三种默认搜索给人看的结果，右侧 Argo 给 Agent 的可吸收证据 JSON"> </p> | 维度 | 模型自带搜索 | AI 搜索（总结型） | 聚合搜索 / 搜索引擎 | **Argo** | |------|------------|-------------------|---------------------|----------| | 结果形态 | 拼好的长文本 | 给人看的总结页 | SERP 链接清单 | **精简 JSON：证据候选 + 可信度分解** | | 垂直问题（行情 / 化学式） | 泛搜网页 | 泛搜再总结 | 泛搜网页 | **直连垂直源，直接给答案** | | 证据可信度 | 无评分 | 无结构化评分 | 无评分 | **selection · absorption · freshness · 共识** | | 重复查询 | 每次都打网 | 每次都打网 | 靠页面缓存 | **双层缓存（内存 + SQLite），热查询约 10ms** | | 成本控制 | 不可控 | 单次贵 | 免费但费事 | **预算模式，免费优先，Key 全可选** | | 多语言 | 随模型走 | 随模型走 | 随引擎走 | **语言检测 + 引擎语言参数 + 多国语言支持** | > 机制上，Argo 把搜索当成一条**证据管线**：语言检测 → 领域路由 → 多引擎召回 → RRF 融合 → 证据快评，交付的是 Agent 可直接排序、可 `fetch` 复核、不撑爆上下文的材料。工具链不用换，需要换的是「搜索结果应该长什么样」——再往下一层，还有和「再包一层搜索 API 的差别」，那是实现细节的对比。 --- ## 这是什么 **Argo 是给 AI Agent 用的多语言搜索基础设施。** 真实检索从来不是「一种语言 + 一个搜索框」：有人问 A 股行情，有人问 World Cup，有人用日文找动画，有人要 IMDb 上的导演信息。Argo 的出发点很朴素——**按领域、按语言、按需求选路**，把问题送到合适的源，而不是一律扫网页标题。联网搜索与本机文件搜索一体可用。 > 产出不是「链接清单」，而是「证据候选 + 可信度分解」。路选对了，证据才站得住。 ### 和「再包一层搜索 API」的差别 |

Read the full READMERepository license: MIT

argo DSH plugin questions

How do I install Argo as a DSH plugin?

Run the command: `dsh plugin --profile web add "github:taxueseek/argo#main&path:packages/dsh-plugin"`. Then restart `dsh web`. This will add 10 MCP tools. You can also use the one-click install script or npx method if you prefer.

Does Argo require an API key?

No, Argo works out of the box with free engines like DuckDuckGo, local_* engines, and many vertical sources. If you configure API keys for certain engines (e.g., Tavily, Exa), the quality may improve, but they are optional. The tool automatically skips engines that require keys you haven't set.

Can Argo search in multiple languages?

Yes, Argo detects the query language and routes to language-friendly engines. It supports Chinese, English, Japanese, Korean, Cyrillic, Thai, Arabic, Hebrew, Greek, Devanagari, and more. It also avoids contaminating non-Chinese queries with Chinese-only sources.

How does Argo's evidence verification work?

Argo tags high-risk queries (finance, medical, legal) with `fetch_required` and each result with `fetch_suggested`. You can use `--verify 3` to automatically fetch the top results, extract the body, and backfill a verified evidence score. Verified URLs are remembered for future searches.

What MCP tools does Argo provide?

Argo provides 10 MCP tools: `argo_search`, `argo_research`, `argo_evidence`, `argo_clarify`, `argo_fetch`, `argo_screenshot`, `argo_pdf`, `argo_social_search`, `argo_local_search`, and `argo_crawl`. These cover unified search, deep research, evidence verification, disambiguation, page fetching, screenshots, PDF extraction, social media search, local file search, and site crawling.