Start here · Quickstart · Skills · Lifecycle · Install · MCP · Contributing
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.
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.
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):
{
"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.
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 a | You say | It 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 test | create-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 month | test-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 coverage | → → → |
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.
| Skill | Stage | What it does |
|---|
| platform-setup | setup | Installs and verifies the Katalon MCP, and diagnoses auth or access failures. Start here. |
| get-started | setup | First-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-plan | 1 plan | Turns 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-estimation | 1 plan | Sizes 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-cases | 2 design | Reads 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-data | 2 design | Designs 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-charter | 2 design | Writes a charter with a mission, areas, oracles, and a timebox, runs the session, and turns what was found into defects and scripted cases. |
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.
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.
npx skills add katalon-labs/true-skills # all skills, this project
npx skills add katalon-labs/true-skills --skill platform-setup # just one
npx skills add katalon-labs/true-skills -g # user-global, every agent
npx 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.
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 plugin marketplace add katalon-labs/true-skills
claude plugin install katalon-true-platform@katalon-true-platform-marketplace
Developing against a checkout:
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 add katalon-labs/true-skills
Open Plugins in Codex and install Katalon True Platform. Set your subdomain in the plugin's .mcp.json.
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.
cp -R <checkout>/.cursor .cursor
Rules in .cursor/rules/*.mdc load by description when relevant. MCP config lives in .cursor/mcp.json.
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 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.
# local/developer install: point Kiro Crew at the app directory
apps/katalon-testing-skills/
Details in apps/katalon-testing-skills/README.md.
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.
cp -R <checkout>/.clinerules .clinerules
Cline loads every file in .clinerules/. Add the MCP through its MCP Servers panel.
cp -R <checkout>/.continue .continue
Rules live in .continue/rules/*.md, the MCP block in .continue/mcpServers/katalon.yaml.
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:
{
"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:
npx -y mcp-remote https://platform.katalon.io/mcp --transport http-first
- Complete the browser OAuth flow that opens on first connect.
- Pick your workspace.
- 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.
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
skills/ source of truth, 19 skills
scripts/build-adapters.mjs generates every agent config
plugins/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:
node scripts/validate-skills.mjs # skills/ matches scripts/skills.config.mjs
node 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:
python3 scripts/build-chatgpt-plugin.py # dist/chatgpt/katalon-true-platform/ and the ZIP
python3 -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.
