Bizintel

io.github.bch1212由 GitHub 帳號 bch1212 發布v0.1.0更新於 Oct 24, 2026

Local business intel for AI agents: audits, lead scoring, tech stack, prospecting.

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取商業與電商

概覽

AI 產生的概覽

遠端 MCP 伺服器，用於在地商家開發：稽核網站、為潛在客戶評分、辨識技術堆疊，並依產業與城市尋找商家。

可以做什麼
稽核單一網站或批次網址的 SSL、行動裝置相容性、速度、聯絡表單與線上預約狀況，並依最差優先排序以便挑選潛在客戶，同時辨識目標網址上的 CMS、預約、郵件與分析工具。依產業與城市搜尋在地商家，取得單一商家的電話、地址、網站、營業時間與評分，並結合稽核與需求訊號給出 0-100 潛在客戶評分。找出沒有網站或沒有線上預約系統的商家，作為冷開發與 SaaS 推銷名單。
適用情境
適合業務開發、潛在客戶開發與競品研究：助理需要依品質排序的在地商家名單、可聯絡的詳細資訊，以及快速的網站品質訊號，而不必人工蒐集。
執行需求
遠端 streamable HTTP 端點，無需本機安裝。需要在 X-API-Key 標頭中提供 API 金鑰。免費開發金鑰每 24 小時限 20 次呼叫；另有付費 Pro 與按次計費方案。商家搜尋在設定後使用 Yelp，否則改用 OSM。
安裝前請注意
超出免費額度後需付費或按次計費。API 金鑰屬於憑證，應避免出現在共用日誌中。稽核會抓取目標網址，商家資料來自第三方來源（Yelp、OSM），結果可能不完整或涵蓋不均。呼叫會依金鑰記錄在持久化帳本中。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Bizintel，將其新增到工作區。
  2. 連接 X-API-Key 憑證。它們會被私密儲存，絕不會在公開頁面上展示。
  3. 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/io.github.bch1212%2Fbizintel/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": {
    "bizintel": {
      "type": "http",
      "url": "https://mcp-bizintel-production.up.railway.app/mcp",
      "headers": {
        "X-API-Key": "<YOUR_API_KEY>"
      }
    }
  }
}

README

BizIntel MCP — Local Business Intelligence for AI Agents

Real-time website audits, lead scoring, tech-stack detection, and local-business search — exposed as an MCP server. Built for AI agents doing sales outreach, competitor research, and prospecting at scale.

MCPize Listing Copy

Title: BizIntel MCP — Real-Time Local Business Intelligence

Subtitle: Audit any website, score leads, find businesses with no booking system. Pay per call.

Description (2 paragraphs):

BizIntel is a paid MCP server that gives AI agents instant access to the kind of local-business intelligence sales teams used to pay analysts to gather. One call to audit_website returns a 0-100 score across SSL, mobile-readiness, page speed, contact form presence, and online booking — plus a tech-stack fingerprint (CMS, booking platform, email provider, analytics). Agents pointed at "find dentists in Austin with no booking system" get a ranked, contactable list in seconds, not hours.

The MCP exposes eight tools — audit_website, bulk_audit, search_businesses, get_business_details, score_lead, find_no_website, find_no_booking, get_tech_stack — backed by Yelp Fusion (or OSM/Overpass when no Yelp key is provided), an aiohttp scanner running 10 concurrent fetches, and a 24-hour SQLite cache so repeat calls don't burn quota. Drop it into Claude Desktop, Cursor, or any agent and your prospecting pipeline becomes one tool call wide.

Pricing

TierPriceLimits
Free / Dev$020 calls / 24h
Pro$19/moUnlimited
Pay-as-you-go$0.05/callNo floor

Upgrade: https://mcpize.com/bizintel-mcp

Tools

ToolArgsReturns
audit_websiteurlScore 0-100, SSL, HTTPS redirect, viewport, load_time_ms, contact form, booking, tech_stack
search_businessesniche, city, state, limitList of normalized business records
get_business_detailsbusiness_name, cityFull record: phone, address, website, hours, rating, lat/lon
bulk_auditurls (≤20)Audit results sorted worst→best (best leads first)
score_leadbusiness_name, city, nicheComposite 0-100 lead score with breakdown
find_no_websiteniche, city, state, limitHottest cold-outreach leads — ranked
find_no_bookingniche, city, state, limitHave a site, no online booking — SaaS-pitch ready
get_tech_stackurlCMS / booking / email / analytics fingerprint

Quickstart

Add to Claude Desktop / Claude Code

bash
claude mcp add bizintel-mcp --url https://mcp-bizintel.up.railway.app/mcp

Set your API key in the MCP config (header X-API-Key). The default dev key bizintel-dev-key-001 is good for 20 calls per day.

Direct HTTP

bash
curl -X POST https://mcp-bizintel.up.railway.app/v1/find_no_booking \  -H "X-API-Key: bizintel-dev-key-001" \  -H "Content-Type: application/json" \  -d '{"niche":"dentist","city":"Austin","state":"TX","limit":10}'

Example agent prompt

Find dentists in Austin with no booking system, audit the top 5, and write a one-line cold-email opener for each that references their actual tech stack.

Local Dev

bash
pip install -r requirements.txtcp .env.example .env  # fill in YELP_API_KEY (optional)python -m uvicorn server:app --reload --port 8000pytest -v

Without a Yelp key, search/details fall back to OSM (Nominatim + Overpass). Coverage is sparser but free.

Deploy

bash deploy.sh from Brett's Mac — pulls secrets from the shared workspace .deploy-secrets.env, links/initializes the Railway project, sets env vars, and runs railway up. Pre-deploy pytest is part of the script so you can't ship a broken build.

Architecture

server.py                FastAPI + fastmcp; HTTP at /v1/* and MCP at /mcptools/audit.py           Async aiohttp auditor (10 concurrent, 5s timeout)tools/techstack.py       CMS / booking / email / analytics fingerprintstools/search.py          Yelp Fusion primary, OSM Overpass fallbacktools/scoring.py         Composite lead score (0-100)db/cache.py              SQLite cache + per-key call ledgerdb/keys.py               Tier classification + 24h sliding windownixpacks.toml            Railway build config

Caching: audits cached 6h, business details 24h, search results 12h. The cache is a single SQLite file with WAL mode — no Redis needed for the price point.

Known Limitations

  1. Yelp doesn't expose external website URL in /businesses/search. We return yelp_url as a stable handle; for a real domain, agents should chain get_business_details (returns hours/photos) and follow the Yelp page or use find_no_website (where OSM-tagged sites are surfaced).
  2. OSM coverage is uneven — some niches map cleanly (dentist, restaurant); long-tail US small business categories (pickleball coach, yacht detailer) won't resolve.
  3. No headless rendering — JS-heavy sites that gate content behind hydration won't expose contact/booking signals to the audit. This is intentional; we trade completeness for 5-second batched audits.
  4. Tech-stack detection is signature-based, not Wappalyzer-grade. We catch the common 90% (WP, Wix, Shopify, Squarespace, Calendly, Mindbody, GA4, Klaviyo) — not obscure custom stacks.
  5. Rate-limit window is sliding 24h, stored per-API-key in SQLite. Restart the container and the ledger persists; clear the DB to reset all dev quotas.

License

Proprietary — © 2026.

來源：README.md，提交 d0617a6

工具

8

audit_website

Score a website 0-100 across SSL, mobile-readiness, speed, contact, booking.

輸入參數url必填

search_businesses

Find local businesses by niche + city. Yelp-backed; OSM fallback if Yelp not configured.

輸入參數city必填limitniche必填state

get_business_details

Resolve a single business to phone, address, website, hours, rating.

輸入參數city必填business_name必填

bulk_audit

Audit up to 20 URLs concurrently. Results sorted worst-first (best leads first).

輸入參數urls必填

score_lead

Composite 0-100 lead score combining audit + demand signal.

輸入參數city必填nichebusiness_name必填

find_no_website

Find businesses in a niche with NO website at all — hottest cold-outreach leads.

輸入參數city必填limitniche必填state

find_no_booking

Businesses with a website but no online booking system. Big opportunity for SaaS pitches.

輸入參數city必填limitniche必填state

get_tech_stack

CMS, booking platform, email provider, analytics detected on a target URL.

輸入參數url必填

版本歷史

1
  1. v0.1.0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