OpenAny MCP 伺服器檢查指南:新增伺服器不等於授權付費 API 呼叫
Inspect the Tools First: Adding an MCP Server Is Not a Paid API Call
OpenAny 的 MCP 伺服器可在不提供 API key 的情況下列出四種工具,但新增伺服器和查看工具清單不等於授權付費呼叫;前三種用於公開目錄,run_api 可執行資源並可能消耗額度。文章示範在 Cursor 不提供 API key 加入伺服器,並建議付費前核對模型或 API ID、價格及計費單位、請求格式和供應商標示。作者只查閲公開資料,沒有登入、安裝 CLI 或執行付費請求。
You can add a remote MCP server, get a green status light, and still have authorized nothing. That is the useful part, if you treat it as a boundary. It is the expensive part if you treat the tool list as a go-ahead to spend.
This is a hands-on pass through OpenAny's public discovery path, then a hard stop before any paid call. I checked the public docs, MCP guide, Skill file, pricing page, terms, and two model listings on 21 November 2026 (Asia/Shanghai). I did not sign in, did not install the CLI, and did not run a paid request.
OpenAny is the B2A (business to agent) marketplace. Give your agent 3,200+ APIs and 168+ AI models with one key and pay-per-call pricing. openany.si
Add the server with no key
The default Cursor recipe on the MCP guide needs no API key:
{
"mcpServers": {
"openany": {
"url": "https://mcp.openany.si/"
}
}
}
Merge it into ~/.cursor/mcp.json (user scope) or .cursor/mcp.json (project scope) and keep any servers you already have. Then open Customize > MCP, enable openany, and reconnect. Saving the file is not a handshake.
Use Streamable HTTP and https://mcp.openany.si/. A browser GET shows documentation; the protocol uses POST. There is no OAuth flow.
Read the tool list like a permission table
The default server exposes four tools:
| Tool | What it does in the official guide | Spends credits? |
|---|---|---|
openany_overview |
Public catalog metadata | No |
search_apis |
Public catalog search; follow next_cursor if present |
No |
get_api |
Public listing details | No |
run_api |
Can execute a resource | It can |
The first three are how you look. run_api is how you might pay. A listed resource is not runtime verification, and a page of search results is not the size of the catalog.
Stop here. Do not call run_api just to "test that the key works." Public discovery does not validate credentials or the balance.
Attach a key only after you have a bounded task
If you later need authenticated calls, Cursor can expand an environment reference. Remote HTTP does not support envFile. Do not paste the key into chat, a prompt, a URL, or a tracked file.
{
"mcpServers": {
"openany": {
"url": "https://mcp.openany.si/",
"headers": {
"Authorization": "Bearer ${env:OPENANY_API_KEY}"
}
}
}
}
Restart the client after the environment is set. Listing tools still does not prove that billing works.
Claude Code and Codex recipes are on the same official page. Browser ChatGPT and Claude are not covered.
Inspect the listing, then decide whether to Run
Copying a published example is free. Clicking Run in a signed-in Playground can incur charges. The terms are explicit about that.
On 21 November 2026, without signing in, I compared two public model pages:
- DeepSeek v4 Flash published a cURL request example and said no response example was configured.
- MiniMax M3 published a request path and a response example, labeled as an example, plus a note to run if you want the actual response.
Before any paid call, check the route's model or API ID, the published rate and billing unit, the request schema, and the provider label. Community or Verified is not the same as an official publisher API. If a list price and a charged price differ, use the receipt. This article has no receipt.
Eligible new accounts can receive $0.50 in signup credits toward supported balance calls. No subscription is required for standard pay-as-you-go access.
A small approved test, not a health-check loop
Official docs show a Chat Completions probe against https://api.openany.si/v1. Use the ID from the listing you just read, keep max_tokens small, and check Usage afterward. I did not run this:
curl --fail-with-body --silent --show-error \
https://api.openany.si/v1/chat/completions \
-H "Authorization: Bearer $OPENANY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "MODEL_ID_FROM_LISTING",
"messages": [{
"role": "user",
"content": "Reply with exactly: OpenAny connected"
}],
"max_tokens": 32
}'
If a call times out, keep the request reference and check its status before you retry. Skills, the CLI and MCP are different surfaces; a skill is not automatically an MCP tool, and skills run in your agent. OpenAny connects buyers and sellers. It does not host the seller's runtime.
Checklist
- [ ] Server added and tools listed with no key
- [ ]
run_apileft unused until a task and budget are approved - [ ] Listing schema, price and billing unit read
- [ ] Key stored in the environment, not in chat or git
- [ ] One bounded call, then a Usage check
Sources, all read on 21 November 2026: openany.si, docs, MCP guide, SKILL.md, pricing, terms.
來源:DEV Community · dev.to