
OpenSDR
io.github.arif-js由 GitHub 帳號 arif-js 發布v1.0.1更新於 Oct 30, 2026
Open-source AI SDR: B2B prospecting, email enrichment, buying signals and safe outreach staging.
概覽
本機 MCP 伺服器,為助理提供業務開發工具組:潛在客戶搜尋、電子郵件補齊、購買訊號、外聯草稿與暫存式行銷活動。
- 可以做什麼
- 依理想客戶輪廓搜尋並篩選聯絡人與公司,並查看各服務商狀態
- 透過電子郵件瀑布流程補齊聯絡人、驗證信箱並補齊公司資料
- 在本機儲存中儲存、列出、匯出潛在客戶並管理拒聯名單
- 偵測購買訊號並研究目標帳戶,每筆訊號附上來源連結
- 產生外聯簡報、建立行銷活動、推送潛在客戶,並在預覽確認後同步 CRM 紀錄
- 適用情境
- 適合 B2B 潛在客戶開發與外聯業務:讓助理尋找聯絡人、驗證信箱、蒐集有來源的購買訊號並撰寫個人化外聯內容,同時把名單與 API 金鑰留在本機。可先以免金鑰的示範模式在虛構資料上試用。
- 執行需求
- 透過 npx 以 stdio 在本機執行,需要 Node.js 22.12 以上版本。選用的服務商金鑰:APOLLO_API_KEY、HUNTER_API_KEY、TAVILY_API_KEY 或 EXA_API_KEY、INSTANTLY_API_KEY、HUBSPOT_ACCESS_TOKEN。設定項包括 OPENSDR_DATA_DIR(預設 ~/.opensdr)、OPENSDR_MAX_CREDITS_PER_CALL 與 OPENSDR_READ_ONLY。需要能連線至所選服務商的網路。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 OpenSDR,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
OpenSDR
The open-source AI SDR for Claude and any MCP client.
Find prospects, verify their emails, spot buying signals, draft grounded outreach and stage campaigns, all from a conversation with your AI assistant, using your own API keys.
[npm] [CI] [License: MIT] [MCP] [OpenSDR MCP server on Glama]
Quick start · Demo · How it works · Docs · Contributing
What it does
OpenSDR is an MCP server that gives your AI assistant the toolkit of a sales development rep (SDR). You describe who you want to reach; your assistant does the legwork:
- 🔎 Find prospects that match your ideal customer profile (Apollo).
- ✉️ Find and verify work emails with a cost-aware waterfall: Apollo → Hunter → free local checks.
- 📈 Spot buying signals (funding, hiring, leadership changes, launches) with a source link for every one (Tavily or Exa).
- ✍️ Draft grounded outreach: every personalized line cites a real, sourced fact.
- 🚀 Stage campaigns and CRM records in Instantly and HubSpot, only after you approve a preview.
Why it's useful
See it in action
A real session in demo mode (no keys, fictional .example companies):
You: Find VP Engineering leaders at B2B software and fintech companies, get their emails, and check for buying signals.
You: Meridian Pay asked us not to contact them. Push the rest into my Instantly campaign.
Full walkthrough → (find → enrich → signals → draft → stage)
Quick start
Requires Node.js 22.12+.
1. Add OpenSDR to your assistant. Claude Desktop → Settings → Developer → Edit Config:
Claude Code: claude mcp add opensdr -- npx -y opensdr-mcp. One-click for Claude Desktop: download opensdr-mcp.mcpb from the latest release and double-click it. Cursor and other clients: see getting started.
2. Try it with no keys. Restart your client and ask: "Find VP Engineering leaders at fintech companies and draft outreach." Or pick the built-in autonomous_sdr_workflow prompt.
3. Add the providers you have. Every key is optional; add them to env:
Also supported: EXA_API_KEY, INSTANTLY_API_KEY, HUBSPOT_ACCESS_TOKEN. Where to get each key and every setting: configuration.
How it works
- Find: your ideal customer profile becomes precise search filters. Search never returns emails, so it's cheap.
- Enrich: each person goes through the email waterfall, stopping at the first verified result, within your credit budget. Every result shows a trail of what was tried and what it cost.
- Research: buying signals give a reason to reach out now, each with a source URL.
- Draft: an outreach brief hands your assistant sourced facts, a sequence framework and writing rules. The assistant writes; you approve.
- Stage: campaign creation, lead enrollment and CRM sync each return a preview and a single-use token, and run only after you confirm.
OpenSDR has no AI model of its own: your assistant does the thinking and writing, and OpenSDR handles data, verification and safe execution. More on the architecture, waterfall and signals →
Tools
Plus 3 built-in prompts (autonomous_sdr_workflow, icp_prospecting, cold_outreach_sequence) and 4 resources (opensdr://providers, opensdr://lists, opensdr://lists/{name}, opensdr://profile). Full reference →
Safety and privacy
- Preview + confirm for every external write, and the confirm is rejected if anything you approved changed in between.
- Campaigns are drafts, never activated, and leads only go into draft or paused campaigns.
- Do-not-contact list honored by every push, sync and export, with no override.
- Read-only mode (
OPENSDR_READ_ONLY=1) and a per-call credit cap. - Your keys and data stay on your machine: no OpenSDR server, no telemetry, keys never logged or stored.
- Never automates LinkedIn: it only drafts notes for you to send.
You're responsible for complying with CAN-SPAM, GDPR, CASL and your providers' terms. Safety, privacy and compliance →
Documentation
Roadmap
- More sequencers: Smartlead, Lemlist
- More CRMs: Salesforce, Pipedrive
- Remote transport (Streamable HTTP) for web-based clients
- Ideal-customer fit scoring and reply tracking
Have a provider you need? Open a provider request.
Contributing
Contributions are welcome, whether you write the code yourself or with an AI coding agent.
- Start here: CONTRIBUTING.md covers setup, checks and the PR process.
- Add a provider: data sources, sequencers and CRMs plug in behind small capability interfaces; follow the add-provider guide.
- Using an AI agent? AGENTS.md holds the architecture rules and safety invariants. Codex, Cursor and Copilot read it automatically; Claude Code and Gemini CLI load it through
CLAUDE.mdandGEMINI.md. - Security issues: please report privately, see SECURITY.md.
License
MIT © Arif
OpenSDR is an independent open-source project, not affiliated with or endorsed by Anthropic, Apollo.io, Hunter, Exa, Tavily, Instantly or HubSpot. Product names are trademarks of their respective owners.
來源:README.md,提交 8a9506e
工具
0版本歷史
2- v1.0.1最新Oct 15, 2026
- v1.0.0Oct 15, 2026
評分
已滿足適用檢查項的 50%
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25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分