Katalon True Platform MCP

io.github.katalon-studio由 GitHub 帳號 katalon-studio 發布v1.0.0更新於 Oct 28, 2026

Katalon True Platform MCP: design tests, run with AI, upload reports, assess release readiness.

已驗證Streamable HTTP可網頁執行效率與工作流開發者工具

概覽

AI 產生的概覽

Katalon True Platform MCP 將助理連接到你的 Katalon 工作區，用來設計測試案例、以 AI 執行測試、上傳報告並判斷版本是否可發布。

可以做什麼
  • 列出專案與儲存庫、讀取需求並追蹤覆蓋率
  • 建立、更新並連結測試案例，管理套件與資料夾
  • 建立手動執行、啟動 Run with AI 工作階段、輪詢並讀取結果
  • 上傳或驗證 Katalon Studio、JUnit XML 與 Playwright 報告
  • 取得品質指標並提交連結 ALM 的缺陷
  • 評估發布就緒度並產出給管理層的摘要
適用情境
適合已使用 Katalon 的測試團隊，讓編碼助理在真實工作區中把需求轉成案例、執行套件、分析失敗原因，並回答某個版本能否發布。README 另附十九個技能，可依角色與階段導引請求。
執行需求
需要 platform.katalon.io（或自有 Katalon 子網域）的遠端 streamable HTTP 端點。首次連線時透過瀏覽器 OAuth 登入並選擇工作區，無需貼上權杖或金鑰。部分用戶端需搭配 mcp-remote 包裝器；選用技能須另外透過 skills CLI 或外掛市集安裝。
安裝前請注意
僅支援瀏覽器 OAuth 驗證；切勿在對話中貼上或提交密碼、API 權杖、Cookie、JWT、MFA 驗證碼或 OAuth 回呼網址。助理可在工作區建立與更新測試案例、套件、執行與缺陷，會寫入真實專案資料。平台操作會將工作區資料傳送給 Katalon，受其隱私政策規範。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Katalon True Platform MCP，將其新增到工作區。
  2. 連接 所需的 憑證。它們會被私密儲存，絕不會在公開頁面上展示。
  3. 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。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.katalon-studio%2Ftestops-mcp/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "katalon-true-platform": {
      "type": "http",
      "url": "https://platform.katalon.io/mcp"
    }
  }
}

README

Ask your coding agent to read a requirement, design the test cases, run them with AI, upload the reports, and tell you whether the release is safe to ship. The skills do the platform work through the Katalon MCP, so the agent operates your real project instead of guessing at it.

[Example agent session: install the skills, analyze requirement CEL-6, design and import cases, run with AI, upload the report, and get a ship decision]

An example session. Your requirement keys, suites, and verdicts come from your own workspace.

Quickstart

1. Install the skills. The skills CLI detects the agent you already run (70+ supported) and drops the skills into its native directory.

bash
npx skills add katalon-labs/true-skills

2. Point it at your Katalon workspace. Add the MCP server to your agent's config file (the install section says which file yours is):

json
{  "mcpServers": {    "katalon-prod-mcp": {      "type": "http",      "url": "https://platform.katalon.io/mcp"    }  }}

First connect opens a browser OAuth flow and lets you pick your workspace. Nothing to fill in, no tokens to paste, no keys to store.

3. Ask for something real.

text
Set up Katalon MCP and verify my projects.Create manual tests from requirement CEL-6 and link them.Which requirements in this sprint have no test coverage?Run that suite with AI and summarize what broke.Generate a Playwright script from test case TC-1042.Upload my Playwright report to Katalon and verify the run.Is release 3.2 safe to ship?

Or hand over the whole chain: "Analyze CEL-6, design and import the cases, build a suite, run it with AI, and tell me if we can ship."

Start where you are

Four jobs. Say the situation in your own words and the agent picks the skill.

You are aYou sayIt starts at
Manual tester
QA analyst, test analyst		I just got requirement CEL-6 and need cases · Run this suite with AI and tell me what broke · Is this failure the app or my testcreate-test-cases → test-data → execute-test → analyze-failures
Unscripted session? exploratory-charter.
Automation tester
SDET, automation engineer		Turn TC-1042 into a Playwright spec · Get my run into the platform · Our suite went flaky this monthtest-case-to-playwright → playwright-execute → test-maintenance
Test lead
QA lead, QE lead		What should we test first this sprint · Is this suite ready for the pipeline · Which requirements have no coveragetest-plan → test-estimation → test-review → test-management
Test manager
QA manager, head of quality		Can we ship 3.2 · How long will this take and with how many people · Build the summary I present upwardrelease-analyze → test-estimation → test-reporting

Not connected yet? Start at platform-setup, which everyone runs once.

Not sure which one you are, or the request spans several? Ask true-platform-testing and it routes by role and by stage. The full map from intent to skill, including what to fall back to for the parts not built yet, is in lifecycle-map.md.

The skills

Nineteen skills, one folder each under skills/. The agent picks the right one from its description, so you rarely name a skill yourself.

SkillStageWhat it does
platform-setupsetupInstalls and verifies the Katalon MCP, and diagnoses auth or access failures. Start here.
get-startedsetupFirst-run onboarding after the connection is installed. Confirms the signed-in account, picks the default project in at most two questions, and opens the Katalon home view. The ChatGPT plugin runs it on install.
test-plan1 planTurns quality goals into scope, ranks the work by requirement coverage and risk, and builds the folder and suite structure that acts as the executable plan.
test-estimation1 planSizes a test cycle from scope, risk, and historical execution data, resources it across the manual and automated lanes, and states the uncertainty instead of inventing precision.
create-test-cases2 designReads a requirement or free text, designs atomic manual cases with ISTQB techniques as reference, skips duplicates, imports only what is missing, and links each case back to the requirement.
test-data2 designDesigns the data a case runs on, decides what must exist before a run, keeps literals out of the step text, and resets state afterwards so the next run starts clean.
exploratory-charter2 designWrites a charter with a mission, areas, oracles, and a timebox, runs the session, and turns what was found into defects and scripted cases.
test-management3 manageOrganizes folders and suites, finds assets at scale, and produces a requirement to case to suite traceability report with coverage percentage and orphans.
test-review4 reviewReviews coverage, case quality, and flakiness before anything enters the pipeline, and returns Approve, Approve with fixes, or Reject plus the specific weak cases.
execute-test5 executeRuns a case, a list, or a suite as a manual run, a Run with AI session, or scheduled automation, then reports pass, fail, and blocked.
upload-report5 executeRuns automation and uploads or verifies Katalon Studio/KRE, JUnit XML, and Playwright reports on the platform.
test-case-to-playwright5 executeConverts manual cases into Playwright TypeScript with Page Object Model and fixtures.
test-case-to-katalon-studio5 executeConverts manual cases into Katalon Studio test cases and Groovy scripts, keeping the .tc and .ts pair GUID-consistent.
playwright-execute5 executeRuns Playwright specs, ships the report with @katalon/playwright-reporter, and verifies the run landed.
analyze-failures6 analyzeSorts failures into product defect, automation defect, and environment noise, clusters them by signature, and files ALM defects for the real bugs.
release-analyze6 analyzeReads coverage, stability, and defect data to return Ready, Ready with risk, or Not ready, with the reasons attached.
test-reporting6 analyzeTurns platform quality data into the few metrics that answer a stakeholder question, trended across releases, as a summary a manager can present upward.
test-maintenance7 maintainFinds flaky and broken cases from stability history, repairs or regenerates them, and hands the refreshed gap list back to planning.
true-platform-testingallThe router and end to end runner. Routes any request to the right stage, or drives the full chain from requirement to ship decision.

Multi-skill playbooks live in combination-recipes.md. Copy-paste prompts and cross-model notes are in prompt-recipes.md.

The lifecycle

Seven stages, each owned by a skill. The loop closes when maintenance feeds its gap list back into the plan.

[The seven-stage testing lifecycle and the skill that owns each stage, with the maintenance gap list feeding back into planning]

platform-setup sits before all seven, connecting the MCP. Every skill states which Katalon MCP tools it uses and where the platform stops. Full stage map: lifecycle-map.md. Every tool in one line each: mcp-tool-index.md. A self-contained visual for humans: docs/lifecycle.html.

Where the boundary is

Each skill names its limits up front so the agent does not promise work the platform cannot do.

Through the MCP: list projects and repositories, read requirements, create and update and link test cases, manage suites and folders, create manual runs, start Run with AI, poll AI sessions, read results, fetch quality metrics, and file ALM-linked defects.

Not through the MCP: creating requirements, creating a formal Test Plan entity, guaranteeing an AI run finishes, or inspecting the live app UI without a browser tool. A named suite plus a release or sprint association stands in for the Test Plan entity.

Some work is a product surface rather than an API call: object capture and resilience design in Studio, custom fields and Git config and governance in the TestOps UI, self-healing and Time Capsule and TrueTest regeneration, and rerun, terminate, and Live Monitor. The full list is in unavailable-capabilities.md.

Install

Every path needs the Katalon MCP configured.

Any agent via the skills CLI
bash
npx skills add katalon-labs/true-skills                          # all skills, this projectnpx skills add katalon-labs/true-skills --skill platform-setup   # just onenpx skills add katalon-labs/true-skills -g                       # user-global, every agentnpx skills update                                                # pull the latest

The CLI installs skill files only. Configure the MCP yourself with the snippet below, or use the Claude Code and Codex plugin paths, which bundle it.

Any agent via the Agent Plugins standard

This repository conforms to the Agent Plugins specification 1.0.0: plugin.json + skills/ + mcp.json at the repo root form a portable plugin any conformant client can load directly, and plugins/katalon-true-platform/ ships the same contract as a standalone plugin directory. Skills follow the Agent Skills format. Point your standard-aware client at either directory; mcp.json already declares the canonical endpoint as a streamable-http server, so there is nothing to fill in.

Claude Code plugin marketplace
bash
claude plugin marketplace add katalon-labs/true-skillsclaude plugin install katalon-true-platform@katalon-true-platform-marketplace

Developing against a checkout:

bash
claude --plugin-dir plugins/katalon-true-platform

The plugin ships the skills and an .mcp.json. If your setup does not auto-load the bundled MCP, add it manually.

Codex plugin marketplace
bash
codex plugin marketplace add katalon-labs/true-skills

Open Plugins in Codex and install Katalon True Platform. Set your subdomain in the plugin's .mcp.json.

GitHub Copilot agent skills + MCP
bash
mkdir -p .github .vscode \  && cp -R <checkout>/.github/skills .github/ \  && cp -R <checkout>/.github/prompts .github/ \  && cp <checkout>/.github/copilot-instructions.md .github/ \  && cp <checkout>/.github/mcp.json .github/ \  && cp <checkout>/.vscode/mcp.json .vscode/
  • VS Code (agent mode) - skills in .github/skills/ load automatically; prompt files stay available as /platform-setup etc. Start the katalon-prod-mcp server when VS Code offers it: it signs you in through the browser OAuth flow and you pick your workspace. Nothing to configure, no token pasting.
  • Copilot CLI - the same skills are auto-discovered (/skills list to see them). The MCP server auto-loads from .github/mcp.json with no edits; or register it interactively with /mcp add (type http, same URL). The CLI does not read .vscode/mcp.json.
  • Copilot coding agent & code review - read .github/skills/ from your repository automatically. Neither surface supports OAuth remote MCP servers yet, so keep Katalon platform operations in VS Code or the CLI for now.
Cursor project rules
bash
cp -R <checkout>/.cursor .cursor

Rules in .cursor/rules/*.mdc load by description when relevant. MCP config lives in .cursor/mcp.json.

Kiro steering docs
bash
cp -R <checkout>/.kiro .kiro

Steering docs use manual inclusion, so reference one in chat with #true-platform-testing. MCP config lives in .kiro/settings/mcp.json.

Kiro Crew app

Kiro Crew installs this as an app — apps/katalon-testing-skills/ carries an app.json manifest that registers all 19 skills and the katalon-prod-mcp server on install. From the dashboard's Apps page, install from the local path, or submit the repo to the Kiro Crew app registry so it appears under Apps → Discover. First MCP connect opens the browser OAuth flow and a workspace picker.

bash
# local/developer install: point Kiro Crew at the app directoryapps/katalon-testing-skills/

Details in apps/katalon-testing-skills/README.md.

Windsurf rules
bash
cp -R <checkout>/.windsurf .windsurf

Rules in .windsurf/rules/*.md trigger on their description. Add the MCP under Settings → MCP, or in ~/.codeium/windsurf/mcp_config.json.

Cline project rules
bash
cp -R <checkout>/.clinerules .clinerules

Cline loads every file in .clinerules/. Add the MCP through its MCP Servers panel.

Continue rules and MCP block
bash
cp -R <checkout>/.continue .continue

Rules live in .continue/rules/*.md, the MCP block in .continue/mcpServers/katalon.yaml.

Anything else via AGENTS.md

Point your agent at AGENTS.md. It indexes every skill and tells the agent to open skills/<name>/SKILL.md. Use the root .mcp.json for the server config.

Connect the Katalon MCP

One endpoint serves every workspace: https://platform.katalon.io/mcp. Sign-in presents a workspace picker, so there is no subdomain to look up. The canonical shape lives in .mcp.json:

json
{  "mcpServers": {    "katalon-prod-mcp": {      "type": "http",      "url": "https://platform.katalon.io/mcp"    }  }}

Agents without native remote-MCP support (Kiro, Continue, Codex) use the mcp-remote wrapper against the same endpoint:

sh
npx -y mcp-remote https://platform.katalon.io/mcp --transport http-first
  1. Complete the browser OAuth flow that opens on first connect.
  2. Pick your workspace.
  3. Reload the agent if the tools do not show up.

On a dedicated Katalon domain, substitute https://<your-sub-domain>.katalon.io/mcp.

Auth is browser OAuth only. Never paste passwords, API tokens, cookies, JWTs, MFA codes, or OAuth callback URLs into chat, and never commit them. The skills enforce this.

How it is built

The skill bodies live once. Everything each agent needs is generated from them, so no adapter can drift.

text
  skills/                      13 SKILL.md files plus references/     |     |  node scripts/build-adapters.mjs      deterministic, checked in CI     v  Claude Code · Codex · Copilot · Cursor · Kiro · Windsurf · Cline · Continue · AGENTS.md     |     |  every adapter points at the same server     v  Katalon MCP  ->  requirements, test cases, suites, Run with AI, results, defects
text
skills/                              source of truth, 19 skillsscripts/build-adapters.mjs           generates every agent configplugins/katalon-true-platform/       Claude Code and Codex plugin      (generated).claude-plugin/  .agents/            plugin marketplaces               (generated).cursor/  .kiro/  .github/           Cursor, Kiro, Copilot             (generated).windsurf/  .clinerules/  .continue/ Windsurf, Cline, Continue         (generated)apps/katalon-testing-skills/         Kiro Crew app                     (generated).mcp.json  .vscode/mcp.json          MCP config                        (generated)AGENTS.md  llms.txt                  agent-readable index              (generated)

Contributing

Edit skills/ only. Everything else is generated. Then:

bash
node scripts/validate-skills.mjs    # skills/ matches scripts/skills.config.mjsnode scripts/build-adapters.mjs     # regenerate every agent config

The build is deterministic. Re-running it with no skill changes produces no diff, and CI rejects out-of-sync adapters. Details in CONTRIBUTING.md.

The ChatGPT plugin package has its own sources under chatgpt/ (the ChatGPT plugin.json, per-skill agents/openai.yaml overlays, the card tool map and the listing assets) and its own build, which stages an allowlisted copy in dist/chatgpt/ and refuses anything the OpenAI plugin directory rejects:

bash
python3 scripts/build-chatgpt-plugin.py      # dist/chatgpt/katalon-true-platform/ and the ZIPpython3 -m unittest discover -s tests        # one bad fixture per forbidden item

Listing copy, review cases and the submission checklist are in docs/chatgpt/.

Support

Privacy

The skills are plain markdown that runs inside the agent you already use. This repository collects nothing. Platform operations go to your own Katalon workspace over OAuth, and what Katalon does with that data is covered by the Katalon Privacy Policy.

License

MIT © Katalon. "Katalon" and "Katalon True Platform" are trademarks of Katalon, Inc.

來源：README.md，提交 334993f

工具

0
該伺服器需要登入，因此 SourceWeft 無法在沒有帳號的情況下讀取其工具，也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v1.0.0最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list，或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分