Gridzen Verification

ai.gridzenv0.2.0更新於 Oct 8, 2026

Country verification research, integration planning and synthetic tests. No live identity checks.

已驗證Streamable HTTP可網頁執行Developer ToolsSecurity & Monitoring

概覽

AI 產生的概覽

讓助理研究各國身分驗證涵蓋情形、規劃驗證流程，並以合成資料進行沙盒測試，不執行真實身分查核。

功能
Gridzen Verification 提供五個工具：get_coverage、plan_verification、create_sandbox_verification、get_sandbox_verification 與 explain_result。它們以內建的離線研究目錄（涵蓋 198 個國家與地區）及合成測試資料運作，因此助理可以查詢某國支援哪些驗證能力、為付款等事件草擬整合方案，並模擬 timeout、not_found 等結果。這些工具標示為唯讀且冪等，且未啟用任何真實服務供應商。
適用情境
適合在評估或設計身分驗證整合時使用，用來取得國家涵蓋研究、整合方案或合成測試結果，之後再串接真實服務供應商。不適用於查核真實個人或授權真實付款。
執行需求
遠端 Streamable HTTP 端點位於服務供應商的開發者路徑下，不需要帳號或 API 金鑰。本機方式為 Python 3.11+ 的 stdio 伺服器，從 GitHub 發行版或網站下載安裝（尚無 PyPI 套件）；CLI 與 MCP 使用內建離線目錄，HTTP 用戶端與範例則會呼叫公開沙盒。遠端使用限制為每個用戶端位址每分鐘 120 次請求，整體每分鐘 600 次。
安裝前請注意
未啟用任何真實服務供應商，也不會查核真實個人；合成比對結果不能用於授權真實使用者或付款，模擬 ID 只是測試資料，不是唯一交易識別碼。切勿將身分證號、銀行帳戶、證件或憑證傳送到沙盒。遠端使用會把文件中列出的參數傳送給服務供應商，本機 stdio 則維持離線。研究證據不等於可商業使用某服務供應商的授權。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Gridzen Verification，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "verification": {
      "type": "http",
      "url": "https://gridzen.ai/developers/mcp"
    }
  }
}

README

Gridzen Developer Kit 0.2.0

Research for 198 countries/territories, a verification integration planner, five synthetic outcomes, an HTTP SDK, local MCP server and three Skills. No live provider is enabled. No real person is verified. The research snapshot is dated 2026-10-07, with source links and file hashes.

中文指南 · Guía en español

Try it

Public playground: https://gridzen.ai/developers/

Download gridzen-developer-kit.zip, extract it, then:

sh
python3 -m venv .venv. .venv/bin/activatepython -m pip install './gridzen-developer-kit[mcp]'gridzen coverage --country ID --capability bank_account_matchgridzen plan --country ID --event payoutgridzen simulate --country ID --capability bank_account_match --scenario timeoutpython gridzen-developer-kit/examples/payout.py

Python 3.11+. CLI and MCP use the bundled offline catalog and fixtures; the Python HTTP client/example calls the public sandbox. Dependencies are downloaded only during installation. API reference: https://gridzen.ai/developers/api/openapi.json .

MCP

Register a local stdio server in your MCP client. Use the absolute path to the installed executable; configuration syntax varies by client. A conventional mcpServers configuration is:

json
{"mcpServers":{"gridzen":{"command":"/absolute/path/to/.venv/bin/gridzen-mcp","args":[]}}}

Tools: get_coverage, plan_verification, create_sandbox_verification, get_sandbox_verification, explain_result. All operate on local research or synthetic fixtures, with read-only/idempotent annotations. Public research/sandbox MCP endpoint: https://gridzen.ai/developers/mcp (Streamable HTTP, no account or API key). It has the same five tools and zero live providers. Remote usage sends the documented arguments to Gridzen; local stdio remains offline. Limits: 120 requests/minute per client address and 600/minute overall, returning HTTP 429 with Retry-After.

Skills

Copy the individual directories from skills/ into your compatible agent's project or user skill directory. Each entrypoint is SKILL.md; preserve its references/ directory. No marketplace listing or automatic registration is implied.

HTTP SDK

python
from gridzen_developer.client import Gridzenclient = Gridzen()plan = client.plan('ID', 'payout')result = client.simulate('ID', 'bank_account_match', 'not_found')assert result['status'] == 'inconclusive'assert result['verified'] is False

Only documented country/event/capability/scenario fields are accepted. Never send identity numbers, bank accounts, documents or credentials to the sandbox. Simulation IDs encode fixtures and are not unique transaction identifiers. A synthetic match cannot authorize real users or payments. Research evidence is distinct from Gridzen availability or commercial permission.

Repository checkout: local service and checks

sh
python -m pip install '.[test,server]'python -m pytest -qpython scripts/check_mcp.pyuvicorn gridzen_developer.server:app --host 127.0.0.1 --port 8030

Production uses an independent container at localhost:8030 proxied only under /developers/. See deploy/compose.yml and deploy/developers.conf. Removing that proxy file and stopping this container rolls back the preview; the existing API and Agent Company stacks are separate.

Repository checkout: build distributions

python scripts/build_downloads.py creates allowlisted developer-kit and Skills ZIPs; it excludes tests, deployment files, private credentials and itself from recursive download packaging. python scripts/build_catalog.py <repo>/docs/research/identity-verification/2026-10-07 refreshes the public research projection. Main repository report files remain the provenance source.

Public distribution

Source: https://github.com/immurray/gridzen-developer-kit . Install the Skills:

sh
npx skills add immurray/gridzen-developer-kit

Install the tagged Python/MCP release from GitHub (Python 3.11+ and Git):

sh
python -m pip install "gridzen-developer-kit[mcp] @ git+https://github.com/immurray/[email protected]"

A PyPI package is not published yet. Use the GitHub release or website download until the release tracker records a verified PyPI URL.

Remote client configuration (conventional JSON; client syntax may vary):

json
{"mcpServers":{"gridzen":{"type":"http","url":"https://gridzen.ai/developers/mcp"}}}

For Claude Code: claude mcp add --transport http gridzen https://gridzen.ai/developers/mcp. For Cursor: add the remote URL in its MCP settings. For portable Agent Plugins, use the packaged plugin.json and mcp.json; the portable transport identifier is streamable-http. Marketplace admission is separate from compatibility.

Original code and Skills: MIT. Research provenance and third-party rights: NOTICE.md. Support: [email protected] .

Marketplace availability

Source, release downloads, the official MCP Registry entry and three Skills.sh pages are public. Docker catalog submission is awaiting review. See the publication ledger for exact status and remaining platform steps.

來源：README.md，提交 f5dd9aa

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v0.2.0最新Oct 8, 2026