Web Search

ai.groundroute由 groundroute.ai 发布官方发布者v0.1.0更新于 Oct 27, 2026

Web search for AI agents — one tool across 6 engines, routed to the cheapest + cached.

已验证Streamable HTTP可网页运行网页搜索与抓取AI 与机器学习

概览

AI 生成的概览

托管式网页搜索 MCP 服务器，提供一个 search 工具，在六个搜索引擎间路由并带缓存与故障转移。

可以做什么
  • 用单个 search 工具执行搜索，可选 auto、web、news、academic、answer、page 等模式
  • 按结果数量、时效性、限定域名、语言和国家筛选与调整结果
  • 获得排序后的结果，含网址、标题、摘要、正文、来源引擎和发布时间，answer 模式还可返回带引用的综合答案
  • 查看每次调用的元数据，如请求 ID、缓存层级、是否降级和计费金额
适用场景
适合需要实时网页、新闻或学术检索、又不想分别接入多家搜索 API 的助手；也适合希望由托管服务处理路由、缓存和故障转移的场景。
运行要求
远程 streamable-HTTP 端点；声明需要 OAuth，README 另说明可用 Authorization Bearer 头传 API 密钥（本地 stdio 服务器用 GROUNDROUTE_API_KEY）。工具发现无需密钥，执行搜索需要。需网络访问。
安装前请注意
搜索按次计费，费用在响应元数据中显示；缓存节省按分成计价，也支持自带密钥。API 密钥属凭据，勿放入共享配置。查询会发送给第三方托管服务，再由其转发给外部搜索引擎。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Web Search，将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储，绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体，或 SourceWeft 对话中。智能体会读取该服务器的安装指南，先向你展示发布者、端点和登录方式，经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/ai.groundroute%2Fweb-search/install.md，并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前，先告诉我它的发布者、端点和登录方式，等我确认。不要把 API 密钥写进命令或我会提交的文件；问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "web-search": {
      "type": "http",
      "url": "https://api.groundroute.ai/mcp"
    }
  }
}

README

[GroundRoute, web search MCP server]

[Glama score] [License: MIT] [Python 3.12+] [MCP compatible] [Smithery]

Give your AI agent web search across 6 engines through one MCP search tool. Hosted. Routed. Cached.

Why GroundRoute

  • One tool, six engines. Serper, Brave, Exa, Tavily, Firecrawl, Perplexity, behind a single search call. Stop wiring up six APIs, six SDKs, six billing portals.
  • Never more than going direct. Gain-share pricing: you keep ~half of every cache saving, GroundRoute keeps ~half. On a miss, you just pay the engine. BYOK supported.
  • Routing, caching, failover, on by default. Each query goes to the cheapest engine that clears a quality bar. Repeats serve from cache. If an engine degrades, we fall back automatically. No agent code changes.

See it work (5 seconds)

A call to the search tool:

json
{  "name": "search",  "arguments": { "query": "what is RAGflow", "max_results": 3 }}

The response (trimmed):

json
{  "results": [    {      "url": "https://ragflow.io/docs/",      "title": "Quickstart - RAGFlow",      "snippet": "RAGFlow is an open-source RAG engine based on deep document understanding...",      "source_engine": "serper"    },    {      "url": "https://github.com/infiniflow/ragflow",      "title": "RAGFlow is a leading open-source Retrieval-Augmented Generation engine",      "snippet": "RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine...",      "source_engine": "serper"    }  ],  "meta": {    "request_id": "req_abc123",    "cache_tier": "miss",    "degraded": false,    "cost_usd": 0.0021  }}

source_engine tells you which engine answered. meta exposes the cache tier and billed cost per call.

Benchmarked, not just shipped

We ran 170 real agent queries across all 6 engines, judged by an LLM, to map cost vs. quality per query class. Full methodology and per-engine results: State of AI Search.

Install

The hosted endpoint is https://api.groundroute.ai/mcp (streamable-HTTP). Get an API key at groundroute.ai/keys.

Claude Desktop / Claude Code, add to your MCP config:

json
{  "mcpServers": {    "groundroute": {      "type": "http",      "url": "https://api.groundroute.ai/mcp",      "headers": { "Authorization": "Bearer gr_YOUR_KEY" }    }  }}

Cursor, ~/.cursor/mcp.json:

json
{ "mcpServers": { "groundroute": { "url": "https://api.groundroute.ai/mcp",  "headers": { "Authorization": "Bearer gr_YOUR_KEY" } } } }

VS Code (native MCP / Continue), .vscode/mcp.json:

json
{ "servers": { "groundroute": { "type": "http", "url": "https://api.groundroute.ai/mcp",  "headers": { "Authorization": "Bearer gr_YOUR_KEY" } } } }

Local / stdio-only clients, bridge stdio to HTTP with mcp-remote:

json
{ "mcpServers": { "groundroute": {  "command": "npx",  "args": ["-y", "mcp-remote", "https://api.groundroute.ai/mcp", "--header", "Authorization:Bearer gr_YOUR_KEY"]} } }

Run this repo's stdio server (optional)

This repo also ships a small native stdio MCP server (server.py) that forwards to the hosted API, useful for stdio-only clients or containerized runs.

bash
pip install -r requirements.txtGROUNDROUTE_API_KEY=gr_YOUR_KEY python server.py

Or with Docker:

bash
docker build -t groundroute-mcp .docker run -i -e GROUNDROUTE_API_KEY=gr_YOUR_KEY groundroute-mcp

Introspection (tool discovery) works with no key; running a search requires GROUNDROUTE_API_KEY (get one at https://groundroute.ai/keys).

The search tool

ParamTypeNotes
querystringrequired
modeenumauto (default), web, news, academic, answer, page
max_resultsintegerdefault 10, max 50
freshnessenumfresh, semi, static; omit to auto-detect
domainsstring[]include-only domain filter, e.g. ["arxiv.org"]
langstringISO 639-1 language code, e.g. en
countrystringISO 3166-1 alpha-2 country code, e.g. us

Returns a structured result: ranked results (url / title / snippet / content / source_engine / published_at), an optional synthesized answer with citations (answer mode), and meta (request_id / cache_tier / degraded / cost_usd). Routed, cached, and reliable.

How it works

One endpoint in front of many search engines, with price-led routing, caching, failover, and usage governance. See the docs and the State of AI Search benchmark (170 real agent queries across all 6 engines).

Links

registry-manifest.json in this repo is the listing manifest for MCP registries.

来源：README.md，提交 0c42389

工具

0
该服务器需要 API 密钥等凭据，因此 SourceWeft 无法在没有凭据的情况下读取其工具，也未收录。

版本历史

1
  1. v0.1.0最新Sep 16, 2026

评分

已满足适用检查项的 85%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分