Hyperresearch Answers
ai.hyperresearch由 hyperresearch.ai 发布官方发布者v1.0.1更新于 Nov 6, 2026
One question, one checked answer: quote-verified answers from the web and ~45 structured sources.
发布方自己的服务：Hyperresearch
概览
远程 MCP 服务器，结合四个网页搜索引擎与约 45 个结构化来源，给出经引文核验的答案。
- 可以做什么
- 提出一个问题，获得一个带引文和链接的已核验答案，耗时较长时返回一个 id
- 按 id 取回仍在处理中的答案
- 搜索并返回排序后的来源，仅含逐字摘录的相关段落，供助手自行阅读证据
- 当来源无法支撑答案时返回 not_found，而不是猜测
- 适用场景
- 适合需要带引用、经核验的研究型答案而非原始网页的助手，尤其是依赖公司申报文件、政府统计、公共登记、学术索引或书籍的问题。也适合希望结果精简、上下文占用小的智能体。
- 运行要求
- 远程 streamable HTTP 端点，无需本地包或运行时。需要 Hyperresearch 账户，首次调用工具时通过 OAuth 登录；无浏览器环境可用 API key，以 Authorization Bearer 请求头发送。调用从账户预付余额扣费，新账户赠送 5 美元额度。
安装
在 SourceWeft 中
- 打开 控制台中的 Hyperresearch Answers，将其添加到工作区。
- 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。
- 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.hyperresearch%2Fanswers/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。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": {
"hyperresearch-answers": {
"type": "http",
"url": "https://answers.hyperresearch.ai/mcp"
}
}
}README
Hyperresearch Answers (MCP)
[Hyperresearch]Every web search provider has gaps. Each one indexes different pages, misses different sources, and none of them reach the structured data behind most real questions.
Hyperresearch Answers layers four web search engines with about 45 structured sources, including company filings, government statistics, public registers, scholarly indexes and books. Your agent gets one search tool with the gaps filled in.
Every answer is checked word for word against quotes from the pages it cites. If the sources don't back it up, it says not_found instead of guessing. Your agent gets the answer and its proof, not whole web pages, so its context stays on your work.
Tools
- ask: one question, one verified answer with its quotes and links. A hard question returns an id instead.
- get_answer: collects an answer that is still being worked on, by its id. Free.
- search: ranked sources with only the relevant passages, copied word for word, for when your agent should read the evidence itself.
Install
Server: https://answers.hyperresearch.ai/mcp (streamable HTTP). You sign in with your Hyperresearch account the first time a tool is called.
Claude Code:
Cline:
Agents setting this up for Cline can follow llms-install.md.
Gemini CLI:
VS Code (GitHub Copilot agent mode):
Any MCP client (mcp.json):
No browser (CI)? Create an API key at https://console.hyperresearch.ai/#/keys and send it as Authorization: Bearer <YOUR_API_KEY>.
Pricing
Calls are paid from prepaid credit in your Hyperresearch account. Every result shows what it cost and the balance left. New accounts get $5 of free credit, no card. Prices: https://hyperresearch.ai/answers
SDKs
Optional packages that give your own agent code the same ask and search tools, over the REST API with an API key (not on npm or PyPI yet):
- Vercel AI SDK:
@hyperresearch/ai-sdk - LangChain (Python):
langchain-hyperresearch - LangChain.js:
@hyperresearch/langchain
Support
[email protected] · https://hyperresearch.ai/answers
The server is hosted by Hyperresearch; nothing here needs to be installed or run to use it. packages/ holds the optional SDKs above.
来源：README.md，提交 67b708d
工具
0版本历史
2- v1.0.1最新Oct 30, 2026
- v1.0.0Oct 26, 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