프로토타입 · 미리 보기: 공식 레지스트리의 모든 서버를 담고 있으며 30분마다 동기화해요. 리뷰와 계정 기능은 다음 단계에서 열려요.
PopNx MCPAI를 쓰는 사람을 위한 MCP 서버 · 쉬운 말로 설명해요

DataCore A2A

제작자: DataCore-VietNam버전 0.2.5i★ –i내 컴퓨터에서 실행
레지스트리 이름vn.datacore/datacore-a2a

설치

개요

제작자가 쓴 설명(영어 원문)

Normalise Vietnamese addresses and look up companies by name or tax code, through DataCore A2A.

PopNx MCP의 세 기둥

이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.

할 수 있는 일

이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.

  • 연결 방식: 내 컴퓨터에서 실행
  • 전송 방식: stdio
  • 카테고리: 회계, 재무 및 세금
  • 입력해야 할 값: DATACORE_A2A_URL, DATACORE_API_KEY, DATACORE_MANDATE

설치

쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.

설치하기 전에

  • 기본 점검: 1개 중 1개 통과
  • 내 컴퓨터에서 내 사용자 계정의 권한(파일, 네트워크, 앱)으로 실행돼요.
  • 직접 입력할 값: DATACORE_A2A_URL, DATACORE_API_KEY, DATACORE_MANDATE, DATACORE_SKILLS, DATACORE_TIMEOUT

비밀로 표시된 값은 내 계정의 열쇠예요. 절대 다른 사람과 공유하거나 채팅에 붙여 넣지 마세요.

필요한 것

  • PyPIPython의 uv 설치(uvx 명령어용)

어떤 앱을 쓰세요?

위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.

Claude Desktop → Settings → Developer → Edit Config를 열고 claude_desktop_config.json에 이 내용을 붙여 넣으세요

{
  "mcpServers": {
    "datacore-a2a": {
      "command": "uvx",
      "args": [
        "datacore-a2a-mcp"
      ],
      "env": {
        "DATACORE_A2A_URL": "your-value",
        "DATACORE_API_KEY": "your-value",
        "DATACORE_MANDATE": "your-value",
        "DATACORE_SKILLS": "your-value",
        "DATACORE_TIMEOUT": "your-value"
      }
    }
  }
}

터미널에서 이 명령어를 실행하세요

claude mcp add datacore-a2a --env DATACORE_A2A_URL=your-value --env DATACORE_API_KEY=your-value --env DATACORE_MANDATE=your-value --env DATACORE_SKILLS=your-value --env DATACORE_TIMEOUT=your-value -- uvx datacore-a2a-mcp

실행하기 전에 빈 값을 채워 주세요.

Cursor에서 열기

또는 설정을 직접 복사하기

프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요

{
  "mcpServers": {
    "datacore-a2a": {
      "command": "uvx",
      "args": [
        "datacore-a2a-mcp"
      ],
      "env": {
        "DATACORE_A2A_URL": "your-value",
        "DATACORE_API_KEY": "your-value",
        "DATACORE_MANDATE": "your-value",
        "DATACORE_SKILLS": "your-value",
        "DATACORE_TIMEOUT": "your-value"
      }
    }
  }
}

브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.

VS Code에서 열기

또는 설정을 직접 복사하기

프로젝트의 .vscode/mcp.json에 붙여 넣으세요

{
  "servers": {
    "datacore-a2a": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "datacore-a2a-mcp"
      ],
      "env": {
        "DATACORE_A2A_URL": "your-value",
        "DATACORE_API_KEY": "your-value",
        "DATACORE_MANDATE": "your-value",
        "DATACORE_SKILLS": "your-value",
        "DATACORE_TIMEOUT": "your-value"
      }
    }
  }
}

브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.

이 서버는 내 컴퓨터에서 실행되는데, ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요. 다른 앱을 고르세요.

선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.

직접 입력할 값

  • DATACORE_A2A_URL 필수 env
    Base URL of the DataCore gateway, for example https://gateway.datacore.vn. It must be the gateway — pointing straight at the orchestrator is always rejected with 401.
  • DATACORE_API_KEY 비밀 env
    Long-lived credential shaped dck_..., created with POST /mcp/credentials — see the README. The server exchanges it for short-lived mandates and refreshes them itself, so it never has to be replaced o…
  • DATACORE_MANDATE 비밀 env
    A hand-issued AP2 mandate, the older way. Not refreshed: when it expires you replace it and restart. Use it only if you were set up this way; otherwise use DATACORE_API_KEY.
  • DATACORE_SKILLS env
    Comma-separated skill ids to narrow the tools that get exposed, or * for every skill the endpoint returns. Empty uses the default four.
  • DATACORE_TIMEOUT env
    Per-request timeout in seconds. Defaults to 30.

이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.

할 수 있는 일

내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.

기본 점검i

누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 6일 점검

  • ✓ 통과

    제작자와 저장소 일치

    레지스트리 이름이 도메인 datacore.vn 기준으로 확인되었어요(레지스트리가 도메인 소유를 확인해요)

  • … 아직 점검 안 함

    계속 관리되고 있음

    아직 GitHub에서 이 저장소를 읽어 오지 않았어요

  • … 아직 점검 안 함

    라이선스 있음

    아직 GitHub에서 이 저장소를 읽어 오지 않았어요

  • – 해당 없음

    엔드포인트 응답

    내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요

  • – 해당 없음

    OAuth로 로그인

    내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요

  • … 아직 점검 안 함

    도구 설명에 숨겨진 지시 없음

    내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요

회계, 재무 및 세금 카테고리의 다른 서버와 비교i

  • 이 카테고리의 다른 서버 1721개 중 0%보다 스타가 많아요
  • 이 카테고리의 다른 서버 1721개 중 78%보다 최근에 레지스트리에 공개됐어요
  • 기본 점검 결과가 이 카테고리의 다른 서버 1721개 중 100%와 같거나 더 좋아요

레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.

자주 묻는 질문

DataCore A2A, 안전한가요?

해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.

어떤 키나 코드가 필요한가요?

입력해야 할 값: DATACORE_A2A_URL, DATACORE_API_KEY, DATACORE_MANDATE

내 컴퓨터에서 실행되나요, 클라우드에서 실행되나요?

내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).

아직 관리되고 있나요?

GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(0.2.5)은 10일 전 레지스트리에 올라왔어요.

어떤 앱에서 쓸 수 있나요?

Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 Python (uv)도 있어야 해요.

대신 쓸 수 있는 건 뭐가 있나요?

같은 카테고리에는: Finance Toolkit, SolidInvoice, Korea Tax Calculator

대안

같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.

카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i

Finance Toolkit

제작자: JerBouma

500+ transparent financial methods calculated from raw statements, not third-party endpoints.

컴퓨터 또는 클라우드★ 5,496MIT어제 업데이트
설치

SolidInvoice

제작자: SolidInvoice

Create and send invoices and quotes, manage clients, and track payments in SolidInvoice.

클라우드에서 실행★ 981MIT7일 전 업데이트
설치

Korea Tax Calculator

제작자: UpstageAI

Korean tax calculation and tax-audit judgment MCP: 96 tools with statutory citations.

내 컴퓨터에서 실행★ 97MIT오늘 업데이트
설치

제작자의 README

README 열기

제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.

DataCore A2A — MCP server

English · Tiếng Việt

Lets an MCP client (Claude Code, Claude Desktop, VS Code, Cursor…) use DataCore's data skills as native tools: normalise Vietnamese addresses, validate an address, look up a company, generate embeddings.

MCP client (stdio) ──► datacore-mcp ──HTTPS──► DataCore gateway

Zero dependencies, Python 3.9+.

Quick start

This server is a client, not a service. It needs a DataCore endpoint and a credential to call it with. Missing either one, it starts, says which variable is absent, and exits.

You need an account on datacore.vn first.

1. Install
pip install datacore-a2a-mcp

If you installed from GitHub while this was unreleased, --upgrade will not

move you to the PyPI build — uninstall first, then install again.

2. Get a session token

Sign in at datacore.vn, then copy the token request header from any call to gateway.datacore.vn (DevTools → Network). 64 hex chars.

No curl equivalent: POST /auth/login RSA-decrypts the password and no endpoint publishes the public key, so only the web app can complete a login. A credential page will remove this step.

Ignore the jwt from the same login — the gateway rejects client-supplied Authorization with 409; token is the header it accepts.

3. Create a credential
curl -s -X POST https://gateway.datacore.vn/mcp/credentials \

-H "token: <64-character token>" \ -H 'Content-Type: application/json' \ -d '{"name":"claude-code"}' | jq .

201 with the credential, including api_key, key_prefix, allowed_skills, spend_cap_vnd and quota_period.

Copy api_key straight away — it is returned once. Only a hash is stored, so there is no way to read it back. Lose it and you create a new one.

A 409 here means you are already holding the maximum of 3 active credentials. Revoke one first (see Quotas and limits).

4. Declare it in your client

See Configuration below.

There is no web UI for steps 2–3 yet; curl is the way. When a management page exists, these commands keep working.

Quotas and limits

Free tier500,000 VND of spend per credential per month
Price per call500 VND by default, so roughly 1,000 calls
Reset1st of the month, Asia/Ho_Chi_Minh
Credentialsat most 3 active per account
Credential lifetime90 days
Mandate issuanceat most 10 exchanges per minute per credential

The cap is counted per credential, not per account — three credentials give an account three separate caps. The service meters in VND rather than in calls so that per-skill prices can be introduced later without rewriting stored caps.

A credential is not bound to a machine or an IP, so the same key works from another computer — on one shared quota, and one revocation kills both.

Check what a credential has spent, and revoke it:

curl -s https://gateway.datacore.vn/mcp/credentials/<id>/usage -H "token: <token>" | jq .

curl -s https://gateway.datacore.vn/mcp/credentials -H "token: " | jq . curl -X DELETE https://gateway.datacore.vn/mcp/credentials/ -H "token: "

/usage reports spent_vnd, spend_cap_vnd, quota_period and resets_at. Revocation returns 204, and 404 if the id is not yours — the two cases are deliberately indistinguishable.

Configuration

VariableRequiredMeaning
DATACORE_A2A_URLyesBase URL of the DataCore gateway — https://gateway.datacore.vn, unless you were given your own endpoint
DATACORE_API_KEYyes¹Long-lived credential, shaped dck_ _ . The server exchanges it for short-lived mandates and refreshes them itself
DATACORE_MANDATEyes¹A hand-issued AP2 mandate. Only for callers set up the old way — see below
DATACORE_SKILLSnoComma-separated skill ids to narrow the tool list, or * for everything the endpoint returns. Empty uses the default four
DATACORE_TIMEOUTnoPer-request timeout in seconds, default 30

¹ One of the two is required. If both are set, DATACORE_MANDATE wins.

Use the URL DataCore gave you. A different endpoint will not recognise your credential and answers 401.

Why DATACORE_API_KEY is the one to use

A mandate lives 15 minutes. With DATACORE_API_KEY, the server requests a fresh one a minute before the current one expires, so you never edit a config file because a token ran out. The mandate stays in process memory and is never written anywhere. If a call still comes back 401, the server forces one refresh and retries exactly once before reporting the failure.

DATACORE_MANDATE still works and is never refreshed — it is there for anyone configured before credentials existed. When it expires you replace it by hand and restart the MCP server.

Claude Code
claude mcp add datacore-a2a -s user \

-e DATACORE_A2A_URL=https://gateway.datacore.vn \ -e DATACORE_API_KEY= \ -- datacore-mcp

Replace with the api_key from step 3 — it looks like dck_<8 hex chars>_ .

-s user registers it for every project. Prefer this over a project-scope .mcp.json: that file tends to get committed, and the key is a 90-day bearer credential.

Claude Desktop

Merge the block below into the config — do not replace the file, it holds your app preferences.

OSConfig path
macOS~/Library/Application Support/Claude/claude_desktop_config.json
Windows%APPDATA%\Claude\claude_desktop_config.json
Linux~/.config/Claude/claude_desktop_config.json
{

"mcpServers": { "datacore-a2a": { "command": " ", "env": { "DATACORE_A2A_URL": "https://gateway.datacore.vn", "DATACORE_API_KEY": " " } } } }

Both <...> values are yours to fill in — JSON has no comments, so nothing in the file will remind you:

PlaceholderWhere it comes from
which datacore-mcp (macOS/Linux) or where datacore-mcp (Windows). Something like /home/you/.local/bin/datacore-mcp or C:\Users\you\AppData\Roaming\Python\Scripts\datacore-mcp.exe
the api_key from step 3, shaped dck_<8 hex chars>_

It has to be an absolute path: the desktop app does not inherit your shell PATH, so a bare command name usually fails to start. python -m datacore_mcp.server works too. Then quit the app completely — closing the window is not enough.

Logs go to stderr, which the app keeps in the logs/ directory beside that config file.

Tools

The tool list is generated from the endpoint, not hardcoded here — so when DataCore adds a skill, restarting the server is enough to see a new tool.

Five skills are exposed by default, the ones wired to a real backend:

ToolSkill idWhat it does
address_normalizeaddress.normalizeNormalises a Vietnamese address into province / district / ward
address_validateaddress.validateChecks an address against administrative data
company_searchcompany.searchLooks a company up by name or tax code
company_graphcompany.graphReturns the company's relationships — representative, province, industry, and anything else the graph holds
company_embeddingscompany.embeddingsGenerates an embedding vector for a piece of text

Every tool takes one required argument, text, a free-text string; no other properties are accepted. Every tool is read-only (readOnlyHint: true): nothing is written, deleted or created, and the same input gives the same result.

DataCore's two other skills (company.financial-sankey, company.news) are deliberately not exposed: they currently return mock/deterministic data. Setting DATACORE_SKILLS="*" lists them as tools, but a free-tier credential cannot call them — its allowlist holds only the five skills above, so every call returns 403. Asking for extra skills when creating a credential is not an error either; the request is silently narrowed to the five.

company.graph used to be on that list. It now reads real relationships, so

a credential created before this release does not include it — the allowed

skills are written into each credential when it is issued. Create a new one to

pick it up.

Examples

Three prompts that go through three different tools:

  1. Normalise an address — "Normalise this address into province/district/ward: So 1 Ly Thai To, Hoan Kiem, Ha Noi"
  2. Company lookup — "Find the company with tax code 0100109106 and tell me whether it is still operating"
  3. Embedding — "Generate an embedding for this line of business and tell me how many dimensions the vector has: wholesale of electronic equipment"

Security model

This server carries the mandate verbatim in the X-AP2-Mandate header for DataCore to verify, and declares no authorization header of its own. A client has no business authorizing itself — it forwards what it was given, and every allow or deny decision stays on the DataCore side.

The API key travels in X-Datacore-Credential when exchanged for a mandate, not in Authorization. DataCore's gateway keeps Authorization for itself — it signs and injects that header after authenticating the caller — so any request that arrives carrying one is rejected with 409 and an empty body. If you call /mandate/issue by hand and get a 409 with no message, that is why.

The API key is a bearer credential: whoever holds it can exchange it for a mandate and call the skills it allows until the cap runs out. Treat it like a password — do not commit a filled-in .mcp.json, and if your client supports it, read the key from a secret manager instead of inlining it. If it leaks, revoke it immediately with DELETE /mcp/credentials/ (see Quotas and limits); mandates already issued from it keep working for the rest of their 15 minutes — there is no instant revocation list.

The mandate is the opposite: 15 minutes of life, process memory only. It is not written to disk, never logged, and you should never need to look at it.

Issuance is rate-limited to 10 exchanges per minute per credential, so the endpoint cannot be used to walk the key space. Every rejection — unknown key, revoked key, expired key, wrong secret — returns the same 401 message for the same reason.

This package never logs a token, not even partially, and never logs the text you send. That is a promise about the process running on your machine — the service side is a separate matter, and PRIVACY.md says plainly what it stores.

Troubleshooting

Logs go to stderr, prefixed [datacore-mcp]. stdout is the JSON-RPC channel.

| S

앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요

PopNx에게 묻기