com.gpuquant由 gpuquant.com 發布官方發布者v1.0.1更新於 Oct 24, 2026
Search AI infrastructure, inspect facility evidence and explore sourced company relationships.
一個託管的唯讀 MCP 伺服器，用來檢索 GPUQuant 有來源的 AI 基礎設施紀錄，並檢視設施與公司的佐證資料。
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。
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.gpuquant%2Finfrastructure/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。把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"gpuquant-ai-infrastructure": {
"type": "http",
"url": "https://gpuquant.com/api/mcp"
}
}
}Connect an AI agent to sourced AI infrastructure records from GPUQuant.
GPUQuant's hosted MCP provides read-only tools for finding infrastructure, inspecting facility facts and checking company relationships. Results link to GPUQuant records and their supporting sources, retaining verification dates and unknown values.
Connection URL: https://gpuquant.com/api/mcp
Transport: Streamable HTTP. No API key is required for this public infrastructure service.
This repository contains connection instructions, the registry manifest and a reproducible HTTP smoke test. The hosted service runs on GPUQuant; cloning this repository does not copy or self-host GPUQuant's database.
In an MCP-compatible application's remote-server settings, add the connection URL above. Use the application's own remote MCP support; configuration format varies by client.
For clients using the VS Code mcp.json format:
See the GPUQuant connection guide and your client's current MCP documentation. The endpoint supports the current MCP protocol and legacy clients using the 2025 initialization handshake.
Use tools/list to obtain the exact current input schemas. Results include canonical GPUQuant URLs so readers can open the underlying record and report a correction.
The first release focuses on infrastructure records and company relationships. It does not expose licensed Ornn or Silicon Data feeds, trading accounts, orders, imagery, building geometry or raw source excerpts. Dates describe the underlying records or sources; a retrieval timestamp is not a fresh verification of a fact.
Announced capacity, construction and operating capacity are different claims. Missing values remain unknown. Supporting sources may describe different dates or scopes; inspect them before relying on a comparison. The MCP supplies research context and does not place trades or provide investment recommendations.
Requires Node.js 20 or newer. No packages or credentials are needed:
The script checks initialization, tool discovery, all three tools against real records, and current-protocol discovery. To test a development endpoint:
The MIT license applies to this repository's original documentation and example code. Data and linked sources retain their own terms. Methodology · Map · Community
來源：README.md，提交 17a7e08
Search infrastructure
Search published facility and company names, slugs, operators, localities, and country codes. Literal text search with bounded pagination; no arbitrary SQL or URL fetching.
kindlimitquery必填offsetGet facility evidence
Read one published facility by slug, including canonical URL, sourced facts, and actual verification date. At most 20 source links; no quotes, imagery, geometry, or market series.
slug必填Get company relationships
Read a company and a page of published supplier, customer, ownership, investment, or facility relationships. Company verification dates are unknown. Each upstream relationship category is capped at 500.
slug必填limitoffsetrelationship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。
{ "servers": { "gpuquant": { "type": "http", "url": "https://gpuquant.com/api/mcp" } }}node scripts/smoke.mjsnode scripts/smoke.mjs http://localhost:3000/api/mcp