Wise MCP Server
Read-only access to Wise (TransferWise) personal API for profiles, balances, exchange rates, transfers, and recipients.
README
@aiwerk/mcp-server-wise
Wise (TransferWise) Personal API MCP server. Read-only: profiles, balances, rates, transfers, recipients.
Tools (v0.1.4, 11 read-only)
wise_list_profiles,wise_get_profilewise_list_balances,wise_get_balancewise_get_exchange_rate,wise_get_exchange_rate_historywise_list_transfers,wise_get_transferwise_list_recipients,wise_get_recipient,wise_get_account_requirements
Write tools (create/fund/cancel transfer, create/delete recipient, create quote) are intentionally not in v0.1.x — they require idempotency + SCA state-machine handling, planned for v0.2.0.
Configuration
| Env var | Required | Default | Purpose |
|---|---|---|---|
WISE_API_TOKEN |
yes | — | Personal API token from https://wise.com/settings/api-tokens |
WISE_API_BASE_URL |
no | https://api.wise.com |
Sandbox override: https://api.sandbox.transferwise.tech |
WISE_API_TIMEOUT_MS |
no | 30000 |
Per-request timeout |
WISE_API_ALLOW_UNSAFE_BASE_URL |
no | — | Dangerous override — see below |
Base URL safety
WISE_API_BASE_URL is locked down by default:
- Must use
https:// - Host must be on the allowlist:
api.wise.comorapi.sandbox.transferwise.tech
Violation throws WiseConfigError before any Bearer token leaves the machine. This prevents token exfiltration via a misconfigured env, a bad CI value, or a malicious override.
Dangerous override: set
WISE_API_ALLOW_UNSAFE_BASE_URL=1to bypass both checks (allowhttp://and arbitrary hosts). Only use for local testing against a mock server. Never enable this in production or with a real Wise token. The token will be sent to whatever host you configure.
Install
Two ways to run this server — pick the one that fits.
Option 1 — Hosted (zero setup)
No local runtime, no env vars on your machine — your Wise token is AES-256-GCM encrypted server-side via HashiCorp Vault.
- Sign up at aiwerkmcp.com.
- Install Wise from the catalog and paste your
WISE_API_TOKEN. - Point your MCP client (Claude.ai, Cursor, Hermes, …) at your hosted endpoint:
with your Bearer token.https://bridge.aiwerk.ch/u/<your-user-id>/mcp
All 11 read-only tools appear. Install other AIWerk recipes from the same bridge.
Option 2 — Self-hosted (npx)
Run directly — you manage the token:
WISE_API_TOKEN=xxx npx @aiwerk/mcp-server-wise
Or install globally:
npm install -g @aiwerk/mcp-server-wise
WISE_API_TOKEN=xxx mcp-server-wise
Error taxonomy
Errors surface as MCP isError: true responses with distinct prefixes:
Timeout: …— request exceededWISE_API_TIMEOUT_MS(wrapsAbortController)Network error: …— fetch rejected (DNS, connection reset, etc.)Configuration error: …— missing token or invalid base URLWise API error <status> …— HTTP 4xx/5xx response from Wise
Typical flow
wise_list_profiles→ pickprofileIdwise_list_balances({profileId})→ see available currencies / amountswise_get_exchange_rate({source, target})→ check ratewise_list_transfers({profileId, status})→ history / status lookup
Build / dev notes
src/version.tsis generated frompackage.jsonbyscripts/gen-version.mjs(runs asprebuild/predev/pretest). The file IS committed so a fresh clone compiles immediately.- The generator is idempotent: it only writes when the version actually changes. So
npm run buildon an in-sync tree leaves the working copy clean. - Bumping
package.json's version and running any ofbuild/dev/testis enough to propagate — no separate manual step onsrc/version.ts.
About AIWerk MCP
Part of the AIWerk MCP platform — curated, signed MCP recipes served either as npm packages for self-hosting or through our multi-tenant hosted bridge (bridge.aiwerk.ch).
Other AIWerk MCP servers:
- @aiwerk/mcp-server-cal — Cal.com scheduling
- @aiwerk/mcp-server-imap — IMAP/SMTP email, provider-agnostic
- @aiwerk/mcp-server-clawhub — ClawHub skill catalog
Browse the full catalog (20+ recipes including GitHub, Linear, Notion, Stripe, …) at aiwerkmcp.com.
Licence
MIT © 2026 AIWerk
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。