LaunchOps
An MCP server that lets AI coding agents autonomously purchase and provision real infrastructure (databases, deployments, domains) with secure, single-use virtual card payments and owner approval gates.
README
🚀 LaunchOps - Autonomous Developer Teammate & B2B Micro-SaaS Procurement Engine
Everything between "I built it" and "it is live."
Powered by Model Context Protocol (MCP) tool orchestration & Prava Virtual Visa Payment Rails ($100 Mandate).
🌐 Live Web App: https://launchops-hackathon.vercel.app/
⚡ The Problem & Solution
Problem
AI coding assistants (Claude Code, OpenCode, Cursor) generate software in seconds. But the moment an agent needs a cloud database, hosting, custom domain, or API usage credits, automation stops dead at the checkout counter:
- No Payment Authority: Giving an AI agent an uncapped corporate credit card risks a $10,000 runaway bill from infinite loops or trial traps.
- Fragmented Context: Withholding payment authority forces developers back into 8 different vendor dashboards to enter credit cards, run SQL migrations, and copy secrets by hand.
Solution
LaunchOps serves as a controlled purchasing layer between coding agents and SaaS vendors. Powered by Model Context Protocol (MCP) tool orchestration and governed by Prava Virtual Visa Cards, LaunchOps turns a half-day manual procurement chore into a single-approval automated launch:
- One-Command Connect: Zero API keys or signup required for judges (
opencode mcp add launchops --url https://launchops-hackathon.vercel.app/api/mcp). - Natural Language Requests: Ask your agent
@launchops setup database for myapp with $15 budget. - Owner Approval Gate: Converts requests into purchase plans requiring biometric Touch ID / Passkey approval before any money moves.
- Single-Use Visa Cards: Prava issues merchant-scoped cards that auto-freeze to $0 balance post-settlement, making runaway charges impossible.
- Real Infrastructure Provisioning: Provisions real Supabase Postgres 16 databases, runs SQL migrations, configures RLS security policies, and automatically writes
DATABASE_URLcredentials to.env.local.
🔌 1-Command MCP Connection (Claude Code / OpenCode / Cursor)
LaunchOps runs a remote WebStandard Streamable HTTP MCP server at https://launchops-hackathon.vercel.app/api/mcp as well as a native local Stdio MCP server (bin/mcp-server.ts).
Quick Add (Remote MCP — Recommended)
Claude Code
claude mcp add launchops --transport http https://launchops-hackathon.vercel.app/api/mcp
OpenCode
opencode mcp add launchops --url https://launchops-hackathon.vercel.app/api/mcp
Cursor
- Settings ➔ MCP ➔ Add Server ➔ URL:
https://launchops-hackathon.vercel.app/api/mcp
💬 4-Step Quick Start Flow (In Your IDE)
Once connected, execute this workflow in your AI assistant chat:
-
Authenticate via GitHub SSO:
LaunchOps loginFollow the returned GitHub OAuth URL to sign in and get your token.
-
Request Infrastructure:
Use LaunchOps to set up a production database for myapp with a $15 budgetLaunchOps generates a Prava payment session and returns a checkout URL.
-
Approve Payment: Open the checkout URL in your browser and use the Prava sandbox test credentials:
- Card Number:
4622943123232291 - Expiry:
12/30| CVV:867| OTP:456789
- Card Number:
-
Confirm & Claim Infrastructure:
Confirm the database for session <session_id>LaunchOps verifies payment, provisions PostgreSQL 16 on Supabase, applies schema/RLS, and syncs connection strings.
🛠️ Exposed MCP Tools
The LaunchOps MCP server exposes 13 tools across remote and local transports:
| Tool Name | Purpose | Inputs |
|---|---|---|
launchops_login |
Day 0 GitHub SSO OAuth authentication & session token check | secretKey?, sessionToken? |
launchops_auth_status |
Returns active vendor connection status & budget summary | (none) |
launchops_setup_database |
Creates Prava payment session & prepares Supabase Postgres 16 DB | domain, appName?, amount? |
launchops_confirm_database |
Polls payment approval & provisions real database, schema & RLS | sessionId, domain, appName? |
launchops_claim_database |
Transfers root ownership & returns connection credentials | sessionId |
launchops_topup_api_credits |
Programmatically tops up API credits (Resend, OpenAI, Anthropic) | vendor, amount? |
launchops_deploy_vercel |
Deploys preview environments to Vercel & syncs secrets | appName, domain? |
launchops_generate_brand |
Synthesizes Recraft V4 Pro SVG logo, favicon & OG cards into ./public/ |
domain, appName? |
launchops_package_devfolio |
Auto-packages Devfolio submission markdown & architecture SVG | domain, appName? |
launchops_buy_domain |
Registers custom TLD via Namecheap/Porkbun API & sets Cloudflare DNS | domain |
launchops_get_budget |
Returns Prava $100 budget governance audit log | (none) |
launchops_get_payment_result |
Queries Prava sandbox transaction settlement status | sessionId |
launchops_report_payment_status |
Reports payment outcome back to Prava network ledger | sessionId, txnRefId, status |
💳 Prava SDK & $100 Budget Governance
All financial operations are governed by Prava Payments (@prava-sdk/core):
- Workspace Cap: Hard limit of $100.00 USD per workspace execution.
- Single-Use Virtual Cards: Generates merchant-scoped Visa virtual cards (
4622943123232291). - Auto-Freeze Protection: Post-settlement, virtual card balances freeze to $0, preventing surprise auto-renewals.
- Enforceable Boundaries: Upgrades exceeding approved session limits fail with:
Prava Authorization Limit Violation: Requested $25.00 exceeds pre-approved single-use card limit ($15.00).
🧪 Local Setup, Testing & Demo App
Clone & Install
git clone https://github.com/piyushagarwal-55/launchops-hackathon.git
cd launchops-hackathon
npm install
Run Vitest Test Suite (40/40 Passing)
npm run test
Quota-Safe Prava Sandbox E2E Harness
# Verify stdio MCP tools
npm run prava:e2e -- list
# Create a single live test session
PRAVA_E2E_CONFIRM=CREATE_ONE_PRAVA_SESSION npm run prava:e2e -- create
Run FeedbackKit Demo App
cd feedbackkit-app
npm install
npm run dev
Open http://localhost:3001 to view the live demo app.
📜 License
Licensed under the MIT License. Prava Payments integration compliant with PCI DSS tokenization standards.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。