Test flows, file bugs and write release notes with an SI agent
Minda Super Intelligence is the AI agent app for your computer, built on Chromium, and for software teams it runs a checkout or sign-up flow with dummy data and files the bug with screenshots, writes release notes from git history, and lets Claude Code, Codex or Cursor read pages over MCP for free, or click and type on Pro.
Coming soon for Windows
- Plan
- Free: MCP read and screenshot; Pro $25 a month adds click and type
- Coding agents
- Claude Code, Codex and Cursor over MCP, off by default
- Bridge
- Listens on 127.0.0.1 only, with a bearer token
- Limit
- Release notes need git installed and read committed history only
Where the hours go in software teams
The job that runs across several apps today, and what Minda takes off it.
- B2B
Software teams
Today
The same sign-up flow clicked through before every release, bugs written up from screenshots, release notes pieced together.
With Minda
The SI agent runs the flow on a schedule and writes the bug report, and coding agents use Minda over MCP.
Every line of work, before and after Seven jobs, step by step
Social posts, supplier prices, lead lists, customer chats, invoices, price drops and research: each app hop and copy-paste of today, beside what Minda does.
See the jobs, before and after
How it works
-
Share the repository, sign in to staging
Share the project folder with the SI agent for reading, and keep staging’s sign-in in a workspace of its own.
-
Describe the job once
Run the checkout with dummy data, stop before payment and file a bug with screenshots if a step fails. Save it as an agent.
-
Put it on a schedule
Schedules run every 15 minutes up to a cron of your own while Minda is open. A job that fails 3 runs in a row pauses itself.
Can an AI agent test a web app’s checkout flow?
Yes: the SI agent walks the flow in a real Chromium tab with dummy data, stops before any payment, and saves screenshots of a failed step for the bug report.
Screenshots are saved as PNG files of the visible part of the page.
Rehearse runs a new saved agent read-only first, so you see its plan before it touches anything.
Each saved agent keeps a record: Proven after 5 clean runs in a row, Slipping when 2 of the last 5 were not clean.
How do I let Claude Code or Cursor use a real browser?
Turn on the developer bridge in Settings, Developer, and Claude Code, Codex or Cursor can open, read and screenshot pages in Minda over MCP.
It is off by default, listens on 127.0.0.1 only with a bearer token, and refuses requests that carry an Origin header. Private windows are never reachable. Reading and screenshots are free; clicking and typing need Pro.
Can AI write release notes from git history?
Yes: it reads the commits since a date in a repository you share, with authors, PR numbers, changed files and API endpoints, and flags likely breaking changes.
Git runs hardened: no system or global config, no hooks, and only log and show between commits.
Author names are read, never email addresses.
Functions and endpoints are found in JavaScript, TypeScript, Python, Go, Java, Kotlin, C#, Ruby and PHP.
The PR description skill writes one from a GitHub or GitLab pull request or compare page.
Pick the closest fit
SaaS product teams
Small product teams who ship every week and watch the flows that earn money
- Runs sign-up and checkout flows with dummy data on a schedule
- Writes weekly release notes from the commits merged since a date
- Watches a provider’s status or pricing page and tells you when it changes
QA and testing
Testers who repeat the same flows and need clear bug reports
- Files a bug with steps and screenshots when a flow fails
- Reads the last lines of a log file and writes an error report
- Opens two builds side by side in split view to compare them
- Added a test item and went to checkout
- Entered the dummy address from the brief
- Saw “Shipping failed”, saved 2 screenshots
- Filed the bug with steps and screenshots
Minda needs your OK
Stopped before payment, as asked
Developer agencies and freelancers
Developers who work across many clients’ repositories, dashboards and staging sites
- Keeps each client’s dashboards and staging sign-ins in a workspace
- Sums up a long issue thread as problem, repro, attempts and status
- Reviews selected code for bugs, security and performance issues, with fixes
Repository and staging site
Issue tracker and docs
Team chat and calendar
Software support teams
Support staff who triage tickets and answer customers about the product
- Triages help desk tickets and leaves them pending with an internal note
- Drafts a support reply to the customer message you select
- Shows new chat and mail messages as cards, with quiet hours in Focus
The features behind it
Each one is part of Minda on its own, and the plan line says which plan includes it.
Developer bridge for coding agents (MCP)
Claude Code, Codex and Cursor can open, read and screenshot pages in Minda over MCP.
PlanFree (read and screenshot), Pro (click and type)
SI agent that acts on pages
Ask the SI to click, type and fill forms on a page for you; it acts on its own and stops for your OK before anything that pays.
PlanPro (Free: 10 actions a day, signed in)
Background, scheduled and scrape tasks
SI briefs and checks that run on a schedule, and no-AI page scrapes that send new rows to a note, file, API, email or WhatsApp.
PlanFree: 1 background and 1 scheduled task. Pro: unlimited
Skills and the community marketplace
Saved prompts, one-click tools and your own page scripts, a gallery of 36 ready-made skills, and community skills to browse, install and share.
PlanFree
Workspaces with separate sign-ins
Each workspace has its own cookies and sign-ins, so you can be in several accounts of the same site at once.
PlanFree: 3. Pro: unlimited
Connectors and integrations
The SI reads your calendar and Notion pages (read-only) and, through MCP, saves to Excel, Word or a Google Sheet and calls your API on its own (Auto mode), each step shown under the reply.
PlanFree
What it does not do
Where this version stops, in plain words, so you know before you rely on it.
Clicking and typing over the MCP bridge needs Pro; on Free, coding agents can open, read and screenshot pages.
The review hold reads a button’s name, so a site that posts publicly from a Save or Comment button is not held.
Schedules run while Minda is open, and a run that needs your OK waits and tells you.
A saved screenshot holds the visible part of the page only.
Frequently asked questions
Is Minda a replacement for Playwright or Selenium?
No. Minda’s SI agent works in a real Chromium browser and you direct it in words, for repeated flows and exploratory checks; it is not a test framework with assertions in code. Coding agents can drive Minda over MCP next to your own tests.
Does Minda send our code to an AI provider?
Only what a request uses, and only to the provider you pick: Minda’s hosted models, your own keys or Ollama on your computer. Keep SI on this computer refuses every cloud provider before anything is sent, and the SI activity log lists each request without its words.
Can the SI agent use our internal tools?
Through MCP, yes: connect an MCP server and the SI agent calls its tools on its own, each step shown under the reply. It also reads your calendar and Notion pages read-only, and can call your own API.
What does a software team pay?
One developer can start on Free. Pro is $25 a month or $240 a year for the agent with no daily limit and clicking over MCP. Team seats are $30 per member a month, with Pro for everyone, team policies and an audit log of every change.
Minda Super Intelligence is coming soon for Windows.
Last updated