Test management for AI agents

TL;DR

Your AI client becomes a QA coworker. Connect Claude, Claude Code, Cursor, or any MCP agent to TestOrchestrator with a personal API key, and it can discover projects, author detailed test cases (steps and priority included), comment, start runs, and link cases to Linear or Jira — all without leaving the tool you already use. The key acts as you: it can only do what you can, in the projects you can access, and never crosses workspaces. Rolling out now.

01The copy-paste tax on AI-assisted QA

Sound familiar?

Your AI assistant drafts great test ideas — edge cases, missing scenarios, a regression list for the feature you just shipped. Then the real work starts: copy each one into your test management tool, reformat the steps, set the priority, file it in the right folder, link it back to the ticket. The intelligence is in one window; the system of record is in another, and you're the integration layer between them.

02Why an MCP connection fits AI-driven teams

The agent writes; TestOrchestrator records

There's no chat UI bolted onto the product. Your own AI client supplies the reasoning; TestOrchestrator supplies the projects, cases, runs, and traceability it acts on — so you keep the assistant you already trust.

Detailed cases, not just titles

Agents can set steps and priority through custom field values, file cases in the right folder, tag them, and link them to a source issue on creation — the same structured cases a person would author by hand.

A key that can only do what you can

An API key inherits exactly your access level, role, project membership, and overrides. Granting an agent access is a least-privilege decision by default — it can never reach a project you can't, or another workspace.

Self-service, no admin ticket

Create a personal key at Profile → API Keys in seconds. It is shown once, stored only as a hash, and you can revoke it instantly. No admin needs to provision anything.

Traceability the agent can read and write

Link a case to a Linear or Jira issue, ask which cases cover an issue, and surface coverage gaps — so the agent closes the loop between work items and tests instead of leaving it to you.

Works with the clients you already run

Claude, Claude Code, and Cursor connect with a one-line config; any MCP-compatible agent works the same way over a standard Streamable-HTTP endpoint.

03Who this is for

Teams that already work with an AI coding assistant — Claude Code, Cursor, or Claude in the loop on a daily basis — and want the test cases, runs, and traceability that fall out of that work to land in a real test management system instead of a chat transcript.

You don't change tools. Your agent gains a set of actions it can take in your QA workspace, on your behalf.

04What it looks like in practice

Connect your client once with a personal API key, then ask it to do QA work in plain language. A single agent can:

Discover the workspace — resolve a project by name, then read its folders, fields, and connected issue trackers.
Author detailed cases — create test cases with steps, priority, tags, and a folder, and update them later — the same structure a person authors by hand.
Comment and review — leave analysis on a case and read what's already there.
Run and read results — start a test run and pull back its pass/fail summary.
Trace coverage — link a case to a Linear or Jira issue, find which cases cover an issue, and surface what isn't covered yet.

The MCP setup guide walks through connecting each client, and the tool reference lists every action with its arguments.

05Why it stays safe

The model that makes this comfortable to turn on is simple: a key acts as its owner. It inherits your permissions exactly and is scoped to your workspace, so handing an agent a key never widens your own reach — if you can't edit a project, neither can the key. Keys are personal, created and revoked by you, shown once, and stored only as a hash.

06Honest limits

Search is keyword/filter, not semantic — it matches on case title and number. The connection is request/response (no server-push channel). And the AI itself lives in your client: TestOrchestrator provides the data and actions, not the model. That's the point — you bring the intelligence you already pay for.

The MCP server is rolling out; if you don't yet see API Keys under your profile, it hasn't reached your workspace yet.

Read the MCP setup guide · Browse the tool reference · Start free

07Frequently asked questions

Does TestOrchestrator have a built-in AI chatbot?

No. There is no chat UI inside the product. You connect your own MCP client — Claude, Claude Code, Cursor, or any MCP-compatible agent — and it calls TestOrchestrator’s tools. The client is the coworker; TestOrchestrator is the system of record.

Can the agent do more than I can in the app?

Never. The API key acts as you — it inherits your access level, global role, project membership, and any per-project overrides, and it can never read or write another workspace’s data. Write tools need the same permissions they need in the app.

Which AI clients can connect?

Any MCP-compatible client. We document one-line setup for Claude Code, Cursor, and Claude Desktop; the endpoint is standard Streamable HTTP, so other MCP agents connect the same way.

What can the agent actually do?

Tools across discovery, test cases (including steps and priority), folders, comments, test runs and results, cycles, and traceability — discover a project’s structure, author and update cases, comment, run and record a test cycle, and link cases to Linear or Jira issues. Search is keyword/filter, not semantic. The tool reference lists every tool.

Test management for AI agents: let Claude or Cursor work in your QA workspace