Katalon MCP server: 40 tools, 18 skills on top

What the forty tools cover, why the vendor ships a written list of things the server cannot do, and which tool its own documentation calls unreliable.

Last verified 25 August 2026 · from The MCP registry, katalon-labs/true-skills and the live endpoint

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Katalon MCP server?

Katalon is a test management platform: requirements come in from Jira or Azure, test cases are written against them, suites run, and results come back with defects attached.

The MCP server exposes that platform to an agent, so a model can read a requirement, design cases against it, run a suite and report whether the release is safe to ship.

The skills are the unusual part. They are markdown instructions, not code, and they encode the order the tools should be called in. The repository states that every skill names which Katalon MCP tools it uses and where the platform stops.

Testing is its whole subject, and it reaches nothing outside it.

For an agent, Katalon answers whether the tests cover the requirement and whether the suite is stable enough to trust. It says nothing about the code under test, or about anything after release.

What can an AI agent do with Katalon?

Fifteen of the 40 tools write, and the vendor's own boundary list decides several rows above. Read that list before promising a task an agent cannot finish.

What connecting costs before the first question

A context window is the amount of text a model can hold at once. Forty tools across the testing lifecycle cost 2,193 to 2,495 tokens on Claude at the hosted endpoint. Katalon publishes a boundary list, which is unusual and worth reading before connecting.

The lifecycle framing is what drives the number. A server organized by test phase carries a tool for each, and all of them arrive together.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted from the 40 tool entries in the vendor's own documentation. That is the only public inventory.

The server could not be run for a live tools/list. So no input schemas are available. These figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint, with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a tenth.

Katalon MCP server limits: where answers come back incomplete

Katalon documents its key limitation explicitly, an unusual level of detail among these vendors.

How to configure Katalon MCP for agents

1. 1

   One endpoint, one picker

   https://platform.katalon.io/mcp serves every workspace, and sign-in presents a workspace chooser. A dedicated Katalon domain substitutes your own subdomain in place of platform.

2. 2

   Decide about the fifteen write tools before connecting

   They create and delete cases, move them between folders, link requirements, file defects and schedule runs.

3. 3

   Read the boundary list

   Eleven capabilities are named as unavailable, including creating requirements, authoring release gates and rerunning or terminating a run.

4. 4

   Install the skills or accept guesswork

   Forty tool descriptions with overlapping names are a lot to choose between. The skills exist to impose an order on them.

5. 5

   Never call delete_test_case from an agent

   Katalon's own index calls it unreliable and recommends update or move instead.

Paste this into your agent’s instructions

The server covers the lifecycle unevenly, and the gaps are documented rather than obvious. The block makes the agent name the boundary instead of inventing a way across it.

When you answer using Katalon MCP tools, state:
- The project and workspace the call acted on.
- Whether the call read or changed something. Fifteen of the
  40 tools write. Ask before any of them.
- When a request falls on the published boundary list, say
  so. Creating requirements, authoring release gates and
  rerunning or terminating a run are not available.
Never state that data does not exist. State which tool you
called and what it returned.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for anything unattended. The Katalon TestOps UI remains the direct route, and a human reading the coverage report is still faster than describing it.

Allure TestOps and ReportPortal cover the same reporting ground with their own servers, and both are cheaper to connect for teams that only need results.

For the test code rather than the test management, Cypress and Playwright answer questions about the run itself. Katalon converts cases into Playwright specs rather than replacing it.

Requirements stay in Jira or Azure either way. Katalon syncs them and reads them, and creating one is on the published exclusion list.

With MCP, an agent can go from a requirement key to a ship decision in one conversation. The 18 skills are written to enforce that sequence.

Connect it for the lifecycle. Read the boundary list before promising anything to a release manager.

Katalon MCP server setup

Nothing is installed for the server itself. The endpoint is https://platform.katalon.io/mcp, and one endpoint serves every workspace.

No configuration block below holds a credential. The first connection opens a browser OAuth flow and then a workspace picker. Katalon's repository states the rule plainly: never paste passwords, API tokens, cookies, JWTs, MFA codes or OAuth callback URLs into chat.

The 18 skills are separate and optional. npx skills add katalon-labs/true-skills installs them into whichever agent directory it detects.

claude plugin marketplace add katalon-labs/true-skills
claude plugin install \
  katalon-true-platform@katalon-true-platform-marketplace

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

[mcp_servers.katalon-prod-mcp]
command = "npx"
args = ["-y", "mcp-remote",
  "https://platform.katalon.io/mcp",
  "--transport", "http-first"]

npx skills add katalon-labs/true-skills
npx skills add katalon-labs/true-skills -g   # every agent
npx skills update

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

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

{"mcpServers":{"katalon-prod-mcp":{
  "httpUrl":"https://platform.katalon.io/mcp"}}}

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

{"mcpServers":{"katalon-prod-mcp":{"serverUrl":
  "https://platform.katalon.io/mcp"}}}

{"context_servers":{"katalon-prod-mcp":{
  "url":"https://platform.katalon.io/mcp"}}}

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

{"amp.mcpServers":{"katalon-prod-mcp":{
  "url":"https://platform.katalon.io/mcp"}}}

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

extensions:
  katalon:
    type: streamable_http
    uri: https://platform.katalon.io/mcp

{"mcpServers":{"katalon-prod-mcp":{"command":"npx",
  "args":["-y","mcp-remote",
    "https://platform.katalon.io/mcp",
    "--transport","http-first"]}}}

{"mcpServers":{"katalon-prod-mcp":{
  "url":"https://platform.katalon.io/mcp"}}}

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

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

name: Katalon
version: 0.0.1
schema: v1
mcpServers:
  - name: katalon-prod-mcp
    command: npx
    args: ["-y", "mcp-remote",
      "https://platform.katalon.io/mcp",
      "--transport", "http-first"]

{"mcpServers":{"katalon-prod-mcp":{
  "url":"https://platform.katalon.io/mcp"}}}

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

npx skills add katalon-labs/true-skills

Read-only access, permissions and security

• Browser OAuth only, and the vendor says so in bold

  The repository instructs readers never to paste passwords, API tokens, cookies, JWTs, MFA codes or OAuth callback URLs into chat, and states that the skills enforce this.

• Fifteen of the forty tools change things

  They create and delete test cases, move them between folders, link and unlink requirements, schedule runs and file defects in a connected ALM.

• create_defect writes into Jira or Azure

  Those are the application lifecycle management (ALM) tools Katalon integrates with, so the defect lands outside Katalon.

• schedule_test_run starts real work

  Scheduled automation consumes TestCloud environments and execution profiles, which cost money and hold state.

• The workspace picker is the boundary

  One endpoint serves every workspace, so the choice made at sign-in decides what the whole session can reach.

Troubleshooting

The tools do not appear after connecting

Reload the agent. Katalon's install steps list that as the third step after completing OAuth and picking a workspace.

The endpoint in an older article is different

There are two registry entries. Version 0.0.1, published 5 December 2025, pointed at product-katalon.katalon.io/mcp. Version 1.0.0, published 10 July 2026 and marked latest, points at platform.katalon.io/mcp.

The agent cannot connect from Kiro, Continue or Codex

Those three have no native remote-MCP support. Katalon documents the mcp-remote wrapper against the same endpoint with --transport http-first.

`delete_test_case` does not delete

Known. Katalon's own tool index annotates it as unreliable and recommends update_test_case or move_test_case instead.

The agent cannot create a requirement

It is on the published boundary list. Requirements come from Jira or Azure and are synced in, so find_requirements and read_requirement are reads only.

A Run with AI session never finishes

Guaranteeing completion is on the boundary list. read_manual_ai_session reports the current state of each item.

Katalon MCP server: Reference

What engineers report

Public discussion of this connector is thin. What exists instead is unusually candid vendor documentation, so the counts below come from it.

“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.”

“Delete a case (unreliable; prefer update/move).”

“Every skill states which Katalon MCP tools it uses and where the platform stops.”

The MCP registry was queried for katalon on 25 August 2026 and returned two entries for io.github.katalon-studio/testops-mcp. Tool names, counts and annotations come from mcp-tool-index.md in the repository that the latest entry names. The endpoint was probed on the same date.

Should you connect an agent to Katalon?

Yes. Best for coverage and readiness questions, with the write tools gated behind approval.

Strongest fit

Katalon TestOps customers who want an agent moving from a requirement key to a ship decision without opening the UI.

Main advantage

The vendor publishes both the tool list and the exclusion list, which removes most of the guesswork an evaluation usually pays for.

Main weakness

Fifteen write tools with no documented read-only mode, and a skills project at 6 stars.

Operational risk

create_defect writes into the connected ALM, and schedule_test_run starts billable automation.

Frequently asked questions