Search

com.neurobird由 neurobird.com 發布官方發布者v1.0.0更新於 Oct 24, 2026

Web search for AI agents. Ranked results with page passages already extracted, plus URL to markdown.

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取知識與記憶

概覽

AI 產生的概覽

託管式網頁搜尋 MCP 伺服器，回傳已擷取頁面段落的排序結果，並能把任意 URL 轉成乾淨的 Markdown。

可以做什麼
搜尋即時網頁，回傳排序結果，可依網域、時間範圍與主題篩選，並可附上根據擷取段落產生的答案，引文會與來源核對。擷取一個或多個 URL，把正文轉成乾淨的 Markdown 或純文字，PDF 會轉為文字。依查詢把擷取到的頁面裁成相關段落。
適用情境
適合助理需要最新網頁資訊且希望直接取得頁面內容、通常不必再抓一次的情況，也適合把指定 URL 或 PDF 轉成乾淨文字以便閱讀與引用。
執行需求
遠端 streamable HTTP 端點，無需在本機安裝。需要 Authorization 標頭攜帶 Bearer API 金鑰；可用免驗證的請求免費建立金鑰，每月 1,000 點，無需註冊或信用卡。需要能連線至該託管服務的網路。
安裝前請注意
API 金鑰透過 Authorization 標頭傳送，請勿放入共用設定。用量以點數計費：進階搜尋與附依據的答案消耗較多，擷取每頁 1 點；免費額度之外另有付費方案。搜尋詞與擷取的 URL 會傳送給第三方服務，結果涵蓋範圍與時效取決於上游搜尋引擎。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Search，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/com.neurobird%2Fsearch/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "search": {
      "type": "http",
      "url": "https://search.neurobird.com/mcp"
    }
  }
}

README

Neurobird Search MCP server

[MCP Registry] [Glama score] [Smithery] [MPPscan] [Transport] [License]

A hosted Model Context Protocol server that gives an AI agent web search with the page content already extracted, plus clean markdown extraction of any URL.

There is nothing to install and nothing to run locally. The server is hosted at https://search.neurobird.com/mcp over the streamable HTTP transport.

bash
claude mcp add --transport http neurobird-search \  https://search.neurobird.com/mcp \  --header "Authorization: Bearer $NEUROBIRD_KEY"

Get a key with one unauthenticated request. No signup form, no email, no card:

bash
curl -X POST https://search.neurobird.com/keys

Every key carries 1,000 free credits a month.

Tools

web_search

Search the live web and get ranked results with the relevant passages already extracted from each page, so a follow up fetch is usually unnecessary.

ParameterTypeDefaultNotes
querystringrequiredThe search query.
search_depthbasic | standard | advancedstandardadvanced expands the query and reads more pages.
max_resultsinteger61 to 20.
include_answerbooleantrueA grounded answer written only from retrieved passages.
include_domainsstring[]Only return results from these domains.
exclude_domainsstring[]
daysintegerRestrict to the last N days.
topicgeneral | news | code | science | financegeneral

extract_url

Fetch one or more URLs and return their main content as clean markdown, with navigation, adverts and boilerplate removed. PDFs are converted to text.

ParameterTypeNotes
urlsstring[]Required.
querystringTrims each page to the passages relevant to it.
formatmarkdown | textDefault markdown.

Other MCP clients

The same URL works in any client that supports the streamable HTTP transport.

json
{  "mcpServers": {    "neurobird-search": {      "type": "http",      "url": "https://search.neurobird.com/mcp",      "headers": { "Authorization": "Bearer nbs-..." }    }  }}

Tested with Claude Code. Cursor, Windsurf, Zed and the Claude desktop app all accept a streamable HTTP MCP server in this shape.

What makes the results different

  • Page passages, not snippets. Each result carries the part of the page that answers the query, extracted from the live page, at no extra credit.
  • Quotes checked against the source. When an answer is requested, every supporting quote is matched back against the extracted source text. A quote that fails is returned marked unverified rather than silently dropped, so an agent can audit a claim instead of trusting it.
  • Rank fusion across engines. Results are fused from several engines with reciprocal rank fusion, so agreement between independent rankings drives the ordering.
  • Per stage timings in every response, so latency can be attributed from outside.

Pricing

Credits: basic or standard search 1, advanced search 2, grounded answer +1, extract_url 1 per page.

Per 1,000 searches
Free tier1,000 credits a month, no card
Basic search$2
Deep search$4
Deep search with grounded answer$6
At 1M credits a month$1 per 1,000

Honest limits

  • Median latency on the current single server deployment is about 15 seconds basic, 22 seconds standard and 28 seconds advanced. Upstream engines and the answer model account for nearly all of it. The HTTP streaming endpoint returns ranked links at about 5 seconds.
  • There is no index of our own. Coverage and freshness inherit from upstream engines.
  • Retrieval capacity, not code, is the scaling limit today.

Links

License

MIT. See LICENSE.

來源：README.md，提交 93cbd02

工具

2

extract_url

Read a URL as markdown

Fetch one or more URLs and return their main content as clean markdown, with navigation, ads and boilerplate removed. Pass a query to get only the passages of each page relevant to it.

輸入參數urls必填queryformat

版本歷史

1
  1. v1.0.0最新Sep 16, 2026

評分

已滿足適用檢查項的 100%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分