# argo [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/taxueseek/argo  
**GitHub Stars:** 226  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/argo

## Description
专门为 agent 打造的开源 agent 搜索工具，具备 200+的搜索源及多语言搜索能力，覆盖中文/英文/学术/代码/购物/金融/新闻/百科。

## Tools
Capabilities this server exposes over MCP:

- **argo_search** — 统一网络搜索：多引擎融合、支持时间过滤与域名过滤，返回带来源与摘要的结果。查资料、找答案、搜新闻/学术/代码/多语言内容等通用场景的首选工具。
- **argo_fetch** — 智能页面抓取：HTTP 优先，遇 Cloudflare/JS 渲染自动升级反检测浏览器。支持 BM25 聚焦提取（focus 省 80%+ token）与结构化提取（mode=extract 抽表格/元数据/JSON-LD）。
- **argo_local_search** — 本机文件/笔记/代码搜索（非联网）：返回文件路径与行号。搜本地资料时用，与网络搜索互补。

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "argo": {
    "command": "npx",
    "args": ["-y","argo-search"]
  }
}
```

## Documentation

## What argo does

argo provides search and research capabilities for agents through MCP, with the same underlying implementation also available as a CLI, skill package, and DeepSeek Harness plugin. Its engine catalog contains 284 sources across 99 business areas and supports 18 languages, including Chinese, English, Japanese, and Korean.

Search can target broad web content or specialized sources for academic material, finance, code, news, entertainment, sports, macroeconomics, and other domains. Results can also include matches from the local machine. API keys may improve the quality or availability of particular sources, but the system can run without them by skipping unavailable authenticated engines and using free or local alternatives.

## How it works

The MCP interface exposes 19 tools that map to the project's search and retrieval pipelines. The main path is ordinary search followed by page fetching when an agent needs evidence or additional context. A two-layer cache can reuse results for repeated queries, while ranking, numerical cross-checking, deduplication, and source safeguards help prepare material for agent consumption.

Four sub-skills organize specialized workflows:

- `local-search` searches filenames, file contents, code symbols, and document structure using local engines. Local matches can be merged into an online search.
- `local-seek` performs hybrid BM25 and vector retrieval over notes, memories, and document collections, progressing from file identification to context and paragraph-level reading.
- `ego-search` uses a real browser and an existing login session to search content behind login requirements, including sites such as Zhihu, Xiaohongshu, and public-account platforms. It is disabled by default.
- `js-run` executes suitable webpage JavaScript without opening a browser, targeting environment checks and pure computation tasks.

The `argo_research` workflow supports single-agent evidence gathering, while `wide_research` can coordinate parallel research tracks, maintain an HTTP(S) source ledger, and apply a quality gate to the resulting report.

## Setup and configuration

Install the npm package globally with `npm install -g argo-search`, run it temporarily with `npx -y argo-search`, or use the GitHub package with `npx -y github:taxueseek/argo`. A source checkout is also supported through Git. The CLI can be tested with `argo search "python asyncio error handling"`, and `argo search --list-engines` lists available engines.

For a persistent MCP connection, attach the server to an MCP client using the npm or local entry point. The project also provides an installation script and injection commands for several clients, including Claude Code, Cursor, Windsurf, Codex, and Cline. No API key is required for basic operation; engine-specific keys can be supplied when higher-quality or authenticated sources are needed.

## Tools and capabilities

- Web search across 284 sources with domain- and language-aware routing
- Web page retrieval for follow-up evidence collection
- Local file and document search with optional web-result merging
- Hybrid semantic retrieval using BM25 and vectors
- Logged-in professional search through a real browser
- Browserless JavaScript execution for supported page scripts
- Caching, TF-IDF reranking, numerical cross-checking, and MinHash deduplication
- Parallel research with source ledgers and quality-gate results

## Limitations and notes

Authenticated engines depend on their configured credentials and may be skipped when keys are unavailable. The logged-in search workflow is opt-in, and browserless JavaScript execution is intended for environment probing or pure computation rather than general browser automation.

The DeepSeek Harness integration offers native tools and a web-search/web-fetch seam in addition to MCP. When both the bundled provider and another provider are available, the web profile must explicitly select the argo search and fetch providers to avoid ambiguous routing. MCP tool definitions can add roughly 3.7K tokens per round for all 19 tools, so the plugin favors native tools for frequent searches and enables the full MCP surface when needed.

_Full upstream README: https://allmcps.com/mcp/argo/readme_