Qiksy Bridge

Give your AI agent the Chrome tab you already have open — your session, your logins. MCP bridge for Claude, Cursor, Codex. Qiksy Bridge is an MCP server for the browser you already have open. Your coding agent — Claude Code, Cursor, Codex, VS Code — reads and drives the tab in front of you: your session, your logins, the state you just broke by hand. No second browser, no sign-in script, no fixtures to maintain.

That is the whole difference from Playwright MCP, which launches a fresh, empty browser and logs in from scratch. Where a flow is already scripted, that one does the job. Where it is not — a client's staging behind SSO, a step you reached by hand, a page you are standing on right now — nothing else can get there.

Install it with npx qiksy-mcp; the popup hands you the ready config for Claude Code, Cursor or .vscode/mcp.json.

And it works with no agent at all. It fills forms with realistic test data, catches console & network errors and accessibility issues while you test, and turns them into paste-ready Markdown bug reports you can drop into any tracker. By default you drive — you click through your app yourself and Qiksy watches every step. Nothing acts on its own initiative; three things act, and each on one deliberate press of yours: Fill puts test data into the form in front of you, a saved scenario replays a form flow you recorded earlier (Pro), and Agent control hands the clicking to your own coding agent — a separate paid feature you switch on yourself, with its own confirmation (see the privacy note below). Qiksy never navigates on its own — only a scenario you recorded and pressed ▶ on can move the tab, and only along the route you walked, including a hand-off to another site; it stops at the first field the page no longer has, and waits for you at every page boundary.

• Smart form-fill — detects fields and fills coherent Valid / Edge / Invalid data in one click, including custom ARIA widgets (date pickers, comboboxes).
• Live findings — console errors, JS exceptions, failed requests, captured as you test.
• Page, flow & interaction audits — labels, contrast, markup, blank/slow routes, dialogs, disclosures.
• Follow mode — flags errors, dead clicks, rage clicks and slow responses tied to the exact click.
• Scenarios (Pro) — record a form flow once, replay it whenever. Press Record, fill the form the way you normally would, press Stop and give it a name; afterwards ▶ replays it, { } lets you edit the run’s data as a plain list of fields, and 🎲 refills them. Group them by project and tags, run a whole project as one set, export them for a colleague. Passwords and card numbers are recorded as fields and never as values.
• Session recorder & client-ready reports — steps, screenshots, HAR.
• Roles & multi-login — save a site login to switch test users, or test the app as admin / member / anon side by side (see the privacy note below).
• Seven languages — the panel and popup speak English, Deutsch, Español, Polski, Português, Русский and Українська, picked from your browser and switchable in the popup. Findings and bug reports stay in English so they can be pasted into a tracker any team can read.
• Copy-as-bug & Bug Capsules — paste-ready Markdown or a self-contained repro page.
• Response mocks (Pro) — answer a request from a rule you write instead of the server (a 500, an empty list, a slow response, a dropped connection), so you can test the failure paths a working backend will not produce on demand. Rules apply to the page's own requests in that tab, are listed in the panel at all times, and are gone when the tab closes.
• Native dialogs (Pro) — decide what the page's own "Leave site? Changes you made may not be saved", confirm(), alert() and prompt() do while you drive a long flow, so an unsaved-changes guard does not stop you on every step. The page still asks: every answer is recorded in Findings together with the question, a panel card names what is being answered with a one-click way to hand the dialogs back, and it applies to that tab only until it closes.

Privacy, plainly: Qiksy runs on your machine. No telemetry, no analytics, no data selling — nothing about the pages you test is ever sent to us. Everything it captures stays in your browser's local storage. Two features can send data off the browser and are off by default, with a consent prompt: the optional Roles / Multi-login features read and locally store this site's cookies (including HttpOnly session cookies) so you can switch logins — this never leaves your device; and the optional local agent bridge (MCP) lets a program you run on 127.0.0.1 read the tabs you put in its session — your findings, the page's structure and rendered text, its computed styles, its localStorage and sessionStorage (often where your session token lives), and captured request URLs and server error bodies — and, only if you additionally switch on Agent control and confirm its own separate prompt, click, type, fill a whole form, attach a file to an upload field and write to that same storage in the app you are testing. That program may forward what it reads to its own AI service. You turn the bridge on in the extension popup, or let the Qiksy web app (qiksy.app) offer to: the site can only ASK — the extension opens its own confirmation page naming the site, the local port and what is being granted, and nothing is enabled unless you accept there. Both are off until you turn them on. One more thing leaves the browser, and only for a paid key: a key opens three installations, so when you paste it — and once a day after — Qiksy asks our licence server whether this installation still holds one of its three places. That request carries the key and a random number the installation made up for itself; no page data, no URLs, no findings, and it is not a device fingerprint. If it fails you keep working. Free and trial users generate no requests at all. Full policy: https://qiksy.app/privacy