@endday/search-mcp 1.0.0

package/README.md

@@ -0,0 +1,117 @@
+# Search MCP
+
+English | [中文](./README.zh.md)
+
+> A local MCP server for aggregated web search and readable content extraction
+
+## What It Is
+
+`@endday/search-mcp` is now a **local-only** MCP package.
+
+It does not ship a remote Worker service anymore. The project is focused on:
+
+- local `web_search`
+- local `content`
+- `jina_content` as an optional reader fallback
+
+## Install
+
+Requirements:
+
+- Node.js 20+
+- an MCP client such as Claude Code, Claude Desktop, OpenClaw, or Codex
+
+Run directly with:
+
+```bash
+npx -y @endday/search-mcp
+```
+
+## MCP Config
+
+Add this to your MCP client config:
+
+```json
+{
+  "mcpServers": {
+    "search-mcp": {
+      "command": "npx",
+      "args": ["-y", "@endday/search-mcp"],
+      "env": {
+        "SEARCH_MCP_CLIENT_ID": "search-mcp-local"
+      }
+    }
+  }
+}
+```
+
+Optional environment variables:
+
+- `SEARCH_MCP_CLIENT_ID`
+- `SEARCH_MCP_UPSTREAM_CLIENT` (`auto`, `impit`, `fetch`)
+- `SEARCH_MCP_PROXY_URL`
+- `SEARCH_MCP_IGNORE_TLS_ERRORS`
+- `JINA_API_KEY`
+- `JINA_BASE_URL`
+- `SUPPORTED_ENGINES`
+- `DEFAULT_ENGINES`
+- `DEFAULT_ENGINES_ZH`
+- `DEFAULT_ENGINES_NON_ZH`
+
+`SEARCH_MCP_UPSTREAM_CLIENT=auto` is the default. In local Node mode it prefers `impit` for upstream requests and falls back to built-in `fetch` if needed. Set it to `impit` to force the impersonated client, or `fetch` to disable it.
+
+## Exposed Tools
+
+- `web_search`: runs local engine fetch + parse and returns ranked results
+- `news_search`: runs explicit news search across supported news-capable engines
+- `content`: fetches a URL and extracts readable text locally
+- `jina_content`: reads a URL through Jina AI reader
+
+## Ranking And Filtering
+
+- blocked domains are filtered locally from generated denylist subscriptions
+- ordinary domains are treated roughly equally by default
+- only a small set of known high-value sources gets deterministic positive boosts
+
+## Supported Engines
+
+- `baidu`
+- `bing`
+- `startpage`
+- `duckduckgo`
+- `brave`
+- `qwant`
+- `yahoo`
+- `mojeek`
+- `toutiao`
+
+Recommended defaults in the current local setup:
+
+- Chinese queries: `baidu`, `bing`
+- Non-Chinese queries: `bing`, `brave`, `yahoo`, `mojeek`
+
+## Dev
+
+Useful commands:
+
+```bash
+npm test
+npm run smoke
+npm run update:blocklist
+npm run docs:dev
+```
+
+## Docs
+
+Long-form docs now live in the VitePress site under [docs-site](./docs-site).
+The docs site is intended to be published with GitHub Pages, so a Cloudflare Worker homepage is no longer needed.
+
+For GitHub Pages:
+
+1. Enable `GitHub Pages` in the repository settings.
+2. Set the source to `GitHub Actions`.
+3. The workflow at `.github/workflows/docs-pages.yml` will build and deploy the VitePress site.
+
+## License
+
+GPL-3.0

package/README.zh.md

@@ -0,0 +1,116 @@
+# Search MCP
+
+[English](./README.md) | 中文
+
+> 一个只做本地 MCP 的聚合搜索与正文提取包
+
+## 现在是什么
+
+`@endday/search-mcp` 现在是一个**纯本地** MCP 包。
+
+仓库已经不再提供远程 Worker 服务，主线只保留：
+
+- 本地 `web_search`
+- 本地 `news_search`
+- 本地 `content`
+- 可选的 `jina_content`
+
+## 安装
+
+要求：
+
+- Node.js 20+
+- 一个支持 MCP 的客户端，例如 Claude Code、Claude Desktop、OpenClaw、Codex
+
+直接运行：
+
+```bash
+npx -y @endday/search-mcp
+```
+
+## MCP 配置
+
+把下面配置加入你的 MCP 客户端：
+
+```json
+{
+  "mcpServers": {
+    "search-mcp": {
+      "command": "npx",
+      "args": ["-y", "@endday/search-mcp"],
+      "env": {
+        "SEARCH_MCP_CLIENT_ID": "search-mcp-local"
+      }
+    }
+  }
+}
+```
+
+可选环境变量：
+
+- `SEARCH_MCP_CLIENT_ID`
+- `SEARCH_MCP_UPSTREAM_CLIENT`（`auto`、`impit`、`fetch`）
+- `SEARCH_MCP_PROXY_URL`
+- `SEARCH_MCP_IGNORE_TLS_ERRORS`
+- `JINA_API_KEY`
+- `JINA_BASE_URL`
+- `SUPPORTED_ENGINES`
+- `DEFAULT_ENGINES`
+- `DEFAULT_ENGINES_ZH`
+- `DEFAULT_ENGINES_NON_ZH`
+
+默认是 `SEARCH_MCP_UPSTREAM_CLIENT=auto`。在本地 Node 模式下，上游请求会优先使用 `impit`，必要时再回退到内置 `fetch`。设为 `impit` 表示强制启用，设为 `fetch` 表示禁用。
+
+## 暴露的工具
+
+- `web_search`：本地抓取搜索结果并返回排序后的结果
+- `news_search`：显式新闻搜索，走支持 news vertical 的引擎
+- `content`：本地抓取网页并提取正文
+- `jina_content`：通过 Jina AI 阅读器读取网页
+
+## 排序与过滤
+
+- 命中本地生成的黑名单域名会直接过滤
+- 普通域名默认大致同权
+- 只有少量已知高价值来源会拿到确定性的正向加分
+
+## 当前支持的引擎
+
+- `baidu`
+- `bing`
+- `startpage`
+- `duckduckgo`
+- `brave`
+- `qwant`
+- `yahoo`
+- `mojeek`
+- `toutiao`
+
+当前本地环境下推荐的默认组合：
+
+- 中文查询：`baidu`、`bing`
+- 非中文查询：`bing`、`brave`、`yahoo`、`mojeek`
+
+## 开发命令
+
+```bash
+npm test
+npm run smoke
+npm run update:blocklist
+npm run docs:dev
+```
+
+## 文档
+
+长文档已经迁到 [docs-site](./docs-site) 里的 VitePress 站点。
+文档站建议直接发布到 GitHub Pages，不再需要单独保留 Cloudflare Worker 首页。
+
+GitHub Pages 配置方式：
+
+1. 在仓库设置里启用 `GitHub Pages`
+2. Source 选择 `GitHub Actions`
+3. 使用 `.github/workflows/docs-pages.yml` 自动构建并发布 VitePress 站点
+
+## 许可证
+
+GPL-3.0