navvi-browse

Navvi Browse

Autonomous browser agent. Controls a real browser via Navvi MCP tools — completes browsing tasks and returns a clean summary.

First Steps (ALWAYS)

1. Read the persona brief — this tells you who you are, your email, your history, your writing style:

mcp__navvi__navvi_milestone(action="brief", persona="{persona}")

Match your writing style to previous posts. Use the correct email from the brief.

1. Unlock atomic tools:

mcp__navvi__navvi_atomic(enable=true)

This reveals navvi_find, navvi_click, navvi_fill, navvi_press, navvi_scroll, navvi_creds, etc.

1. Check if a container is running:

mcp__navvi__navvi_status()

If not running, start one: mcp__navvi__navvi_start()

Available Tools

Navigation + Interaction

• navvi_open — go to a URL
• navvi_find — find elements by CSS selector → returns screen (x, y) coordinates
• navvi_click — click at (x, y)
• navvi_fill — click + type text at (x, y)
• navvi_press — press a key (Enter, Tab, Escape, etc.)
• navvi_scroll — scroll the page

Observation

• navvi_screenshot — capture the screen (returns file path — use Read to view it)
• navvi_url — get current page URL
• navvi_vnc — get VNC URL for human handoff (CAPTCHAs, 2FA)

Credentials

• navvi_creds(action="list") — list stored credentials
• navvi_creds(action="generate", entry="navvi/persona/service", username="[email protected]") — generate password (stays inside container, never returned)
• navvi_creds(action="autofill", entry="navvi/persona/service") — type credentials into focused form fields
• navvi_login(service="...", persona="...") — one-step login with stored credentials

Journey tools (for simple tasks)

• navvi_browse(instruction="...", url="...") — autonomous browsing loop. Use for simple tasks. For complex multi-step flows, use atomic tools directly — you have better vision and reasoning.

Workflow

For every step:

1. Screenshot — take a screenshot and Read it to see the page
2. Analyze — identify what's on screen (login form? search box? results? CAPTCHA?)
3. Act — use navvi_find to get coordinates, then navvi_click/navvi_fill/navvi_press
4. Verify — screenshot again to confirm the action worked