Invoice-MCP
MCP server that gives AI agents read access to subscription/SaaS invoice history, ingested from Gmail and stored in Postgres and MongoDB, with tools for listing invoices, retrieving PDFs, and summarizing spend.
README
InvoiceMCP
MCP server giving AI agents read access to subscription/SaaS invoice history,
ingested from Gmail and normalized into Postgres (structured data) + MongoDB
(PDFs). See SPEC.md for the full design.
Prerequisites
- Node 22+, pnpm
- Docker (local Postgres/Mongo/Redis) — or Railway database URLs
- Google OAuth client (Gmail API enabled, redirect URI
http://localhost:8787/oauth/callback) - OpenRouter API key
Setup
cp .env.example .envand fill in values (scripts/has key generators noted inline)docker compose up -dpnpm installpnpm migrate up— appliesmigrations/to Postgrespnpm cli seed-org --name "Your Org"pnpm cli auth— browser OAuth consent, stores encrypted refresh tokenpnpm cli check-connections— sanity check all three databases
Running
| Command | What |
|---|---|
pnpm worker |
ingestion worker (BullMQ; registers the monthly sync cron) |
pnpm mcp |
MCP server on :3000 (Streamable HTTP at POST /mcp, API key auth) |
pnpm cli discover |
whole-mailbox billing-sender discovery → auto-backfill |
pnpm cli backfill --service <id> |
full history for one sender |
pnpm cli sync |
incremental sync now (otherwise monthly via SYNC_CRON) |
MCP tools
list_services, list_invoices, get_invoice, get_invoice_pdf,
spend_summary — all scoped to DEFAULT_ORG_ID; money in minor units,
per-currency aggregation, no FX conversion.
Claude Desktop's connector UI expects OAuth, which this MVP doesn't
implement — bridge with mcp-remote
in claude_desktop_config.json instead:
{
"mcpServers": {
"invoice-mcp": {
"command": "npx",
"args": [
"-y", "mcp-remote", "https://<host>/mcp",
"--header", "Authorization: Bearer <MCP_API_KEY>"
]
}
}
}
Development
pnpm test/pnpm typecheck/pnpm lintpnpm codegen— regeneratesrc/db/types.tsafter schema changestsx scripts/smoke-extract.ts [pdf]— live LLM extraction smoke testtsx scripts/make-fixture-pdf.ts— regenerate the multi-page fixture PDF
Deploy (Railway)
One repo, two services:
- worker — start
pnpm start:worker - mcp — start
pnpm start:mcp(health check/healthz)
Both build with pnpm install && pnpm build. Run pnpm migrate up as the
worker's pre-deploy command. Set env vars per .env.example (Railway injects
database URLs for linked Postgres/Mongo/Redis services).
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。