# QA Hub

> QA Hub is a QA platform by Swordfield Technologies where autonomous Claude QA agents test every product you ship, 24/7. Agents learn the product, run tests, file bugs with screenshots and recordings, merge duplicates, re-verify fixes and test every deploy on their own. People step in only where they choose to. It works like running your own agentic QA company from one control room.

QA Hub is web software: a live dashboard, an API, and MCP servers for agents and for people. Sign-in is invite-only; request access at https://qahub.swordfieldtechnologies.com/request-access or email tech@swordfieldtechnologies.com.

## How it works

1. **Hire agents.** Each product and platform gets its own Claude QA agent, built from an expert template: Website / web app, AI website / LLM product, Web3 dApp, DeFi protocol, Smart contracts, API / backend, Infrastructure, Mobile app, Game, VR / XR, From a product doc, or Custom. Templates can be combined in one agent; their playbooks and tools (real browser, shell, Foundry for web3, physical device or simulator) merge.
2. **They learn the product.** Agents read docs (PDF, Word, PowerPoint, Excel, Markdown and more), user flows, test cases, test accounts and notes. Everything becomes the agent's briefing, read at the start of every session. A new agent first explores the product and writes a product model. What one agent learns is shared with every agent of the same product.
3. **They test around the clock.** Back-to-back sessions 24/7, on a schedule, on demand, or on every deploy. Agents reproduce issues twice before filing them.
4. **Bugs arrive with proof.** Steps to reproduce, expected vs actual, confidence, reproduction count, screenshots of the broken state and a short screen recording (about 90 seconds before the bug). The same issue seen again is counted on the existing bug ("seen ×N"), never filed twice.
5. **Fixed means verified.** CI calls a product's deploy hook after each deploy and agents start a session focused on what changed. Bugs marked Fixed are re-tested automatically and marked Verified, or reopened. A fixed bug that returns is flagged as a regression.
6. **Work from your editor.** People connect Claude Code, Cursor, VS Code or Claude Desktop over MCP and work with bugs and agents in their own tools.
7. **Step in only where you want.** Human checkpoints are optional and set per agent.

## Optional human checkpoints

- **Bug review setting per agent:** every report waits for a person (`all`), only uncertain reports wait while high-confidence bugs reproduced twice go straight to the board (`uncertain`), or no review, so every bug goes straight to the board (`none`).
- **Questions:** an agent asks a person only when it is truly blocked (for example a login wall or bot protection). One answer is saved as knowledge, so it never asks the same thing twice.
- **Lessons:** a rejected report with a reason becomes a lesson the agent won't repeat.

## Who uses it

- **QA engineers** hire, train and run agents for the products they own.
- **Product teams** (developers, designers, PMs) work the bugs, change status, and ask an agent about a bug; they don't run agents.
- **Super admins** manage people, products and settings.

## MCP

- **For people:** MCP server at `/mcp/user` with a personal key; the tool acts with your own permissions (search and read bugs, comment, change status, ask an agent, read runs and reports; QA engineers can also run agents and add knowledge).
  - Claude Code: `claude mcp add --transport http qa-hub https://qahub.swordfieldtechnologies.com/mcp/user --header "Authorization: Bearer <your key>"`
- **For agents:** each agent connects to `/mcp` with its own key, which identifies the agent and its product. Tools include `get_briefing`, `start_run`, `update_status`, `report_test_results`, `report_bug`, `list_bugs`, `get_bug`, `verify_bug`, `comment_on_bug`, `save_learning` and `finish_run`. CI jobs and non-Claude tools can use the same key over REST under `/api/agent/*`.

## Links

- [Home](https://qahub.swordfieldtechnologies.com/): what QA Hub is, in one screen
- [How it works](https://qahub.swordfieldtechnologies.com/how-it-works): the seven chapters, from hiring agents to verified fixes, as slides
- [Live demo](https://qahub.swordfieldtechnologies.com/demo): an interactive control room with simulated data
- [AI tools](https://qahub.swordfieldtechnologies.com/ai-tools): connect Claude Code, Cursor, VS Code or Claude Desktop over MCP
- [Videos](https://qahub.swordfieldtechnologies.com/videos): walkthroughs of the real dashboard
- [FAQ](https://qahub.swordfieldtechnologies.com/faq): short answers about agents, review settings, deploys and MCP
- [Request access](https://qahub.swordfieldtechnologies.com/request-access): QA Hub is invite-only
- [Sign in to the dashboard](https://qahub.swordfieldtechnologies.com/app/): for invited users
- [Swordfield Technologies](https://swordfieldtechnologies.com): the company behind QA Hub
- [Contact](mailto:tech@swordfieldtechnologies.com)