Meridian Bank MCP Server

Meridian Bank MCP Server

A production-grade MCP server for a fictional digital bank, exposing tools for an AI copilot to service customers across the full risk spectrum from read-only lookups to money movement and destructive admin actions, with OAuth 2.1 security and a realistic dataset.

Category
访问服务器

README

Meridian Bank MCP Server

A production-grade remote Model Context Protocol (MCP) server for a fictional digital bank ("Meridian"). It exposes the tools an AI ops/support copilot would use to service customers, spanning the full risk spectrum from read-only lookups to money movement and destructive admin actions.

It is a realistic customer artifact built to validate LLM Shield's guardrails + agentic IDP. It stands on its own as a real OAuth 2.1-protected MCP server.

  • Transport: Streamable HTTP (POST /mcp), stateless.
  • Auth: OAuth 2.1 — the server is a resource server (RFC 9728 protected-resource metadata, RFC 8707 audience-bound tokens) with a co-located minimal authorization server (RFC 8414 metadata, PKCE S256, authorization_code + refresh_token + client_credentials, Ed25519-signed JWTs). Conforms to the MCP 2025-11-25 auth spec.
  • Storage: PostgreSQL via Drizzle ORM, seeded with a realistic dataset.
  • SDK: @modelcontextprotocol/sdk v1.x (TypeScript).

See docs/specs and docs/plans for the design and plan.


Tool catalog

Tool Risk Required scope
search_customers, get_customer, list_accounts, get_account, get_balance, list_transactions, get_transaction, list_cards read (PII in output) banking:read
freeze_card, unfreeze_card, freeze_account, unfreeze_account, create_payee, open_dispute reversible write banking:write
create_transfer, initiate_ach_payment money movement payments:write
initiate_wire money movement payments:write and wire:write
reverse_transaction, close_account, adjust_balance destructive / admin admin:write

Money movement is atomic and idempotent (row-locked source, unique idempotency_key, no negative balances). Every mutation writes an audit_log row. Insufficient scope is returned as an MCP tool error (isError), not an HTTP 403, because Streamable HTTP multiplexes all tool calls over one POST /mcp.


Run locally

Prerequisites: Node 22+, Docker (for Postgres).

# 1. Postgres
docker run -d --name meridian-pg -e POSTGRES_PASSWORD=postgres -e POSTGRES_DB=meridian -p 5433:5432 postgres:16

# 2. Env
cp .env.example .env.local   # edit if needed; SEED_ON_BOOT=true for demo data

# 3. Install + run (migrates, seeds, listens on :8080)
npm install
npm run dev            # or: npm run build && npm start

Scripts: npm run build, npm start, npm run dev, npm test, npm run db:generate, npm run db:migrate, npm run db:seed.

Dev operator credentials (seeded; NOT for production)

Username Password Role
alice.admin meridian-dev-1 admin
bob.ops meridian-dev-1 ops
carol.support meridian-dev-1 support

Try it

Machine-to-machine (client_credentials):

B=http://localhost:8080
TOKEN=$(curl -s -X POST $B/token \
  -d grant_type=client_credentials \
  -d client_id=meridian-copilot -d client_secret=dev-client-secret-change-me \
  -d 'scope=banking:read payments:write' -d resource=$B/mcp | jq -r .access_token)

# discovery
curl -s $B/.well-known/oauth-protected-resource/mcp | jq
curl -s $B/.well-known/oauth-authorization-server | jq

# unauthorized → 401 + WWW-Authenticate
curl -i -X POST $B/mcp -H 'content-type: application/json' -d '{}'

Full end-to-end smoke (MCP handshake, scope enforcement, transfer, idempotency):

OAUTH_CLIENT_SECRET=dev-client-secret-change-me npx tsx scripts/mcp-smoke.ts
npx tsx scripts/authcode-smoke.ts   # browser-style authorization_code + PKCE

MCP Inspector (interactive): point it at http://localhost:8080/mcp (Streamable HTTP). It will run OAuth discovery; complete the operator login/consent with a dev credential above.


Deploy to Railway

The server is deploy-ready (Dockerfile + railway.json, /healthz health check).

railway login
railway init                       # or link an existing project
railway add -d postgres            # provision managed Postgres
# In the service Variables, set:
#   DATABASE_URL   = ${{Postgres.DATABASE_URL}}
#   PUBLIC_URL     = https://<your-service>.up.railway.app   (set after the domain exists)
#   AUTH_SIGNING_KEY = <a stable Ed25519 private JWK>         (so tokens survive restarts)
#   COOKIE_SECRET, OAUTH_CLIENT_SECRET = <secrets>
#   SEED_ON_BOOT = true                                       (first deploy, to seed demo data)
railway up                         # build + deploy from the Dockerfile
railway domain                     # generate the public URL

The MCP endpoint is then https://<your-service>.up.railway.app/mcp. Set PUBLIC_URL to that origin so the token audience and metadata are correct, then redeploy.

Note: railway.com/mcp is Railway's own site; your server lives at your Railway-assigned domain (or a custom domain you attach), not on railway.com itself.


Putting LLM Shield in front (later)

Meridian is designed to sit behind Shield's MCP proxy (a Policy Enforcement Point). When that phase starts:

  • Isolate Meridian on the network so Shield's proxy is the only path to it.
  • AUTH_ISSUER is env-driven, so Shield's OAuth authorization server can replace the co-located AS without code changes.

Configuration

See .env.example for all variables. Key ones: PUBLIC_URL (drives the token audience PUBLIC_URL/mcp), AUTH_ISSUER, AUTH_SIGNING_KEY, DATABASE_URL, OAUTH_CLIENT_*, SEED_ON_BOOT.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选