Give your agent the whole bug.
snagisk captures what happens in the browser, with the console logs and network requests attached. Agents read bug reports over MCP, or read and capture them with the CLI.
For agents
Your agent sees what the browser saw.
Connect it through the MCP server or the CLI. You only need one.
Read snags
Your agent fetches any snag you send it, with the video and everything logged alongside. There's nothing to install.
Pick this if people capture the bugs and your agent fixes them.
{
"mcpServers": {
"snagisk": {
"type": "http",
"url": "https://snagisk.example/mcp"
}
}
}Read and capture snags
Everything the MCP server does, plus capturing new snags while your agent works in the browser.
Pick this if you want your agent to show its work.
$ curl -fsSL https://snagisk.example/install.sh | shCLI
Then just ask
Tell your agent what to check. It captures while it works and sends you a snag as proof.
> Check that checkout works on localhost:3000 and send me a snag. ⏺ Started a snag on localhost:3000/checkout⏺ Filled in the card form and clicked Pay⏺ POST /api/checkout returned 500: cart.total is undefined⏺ Fixed the total in cart.ts and tried again. Payment went through. ✓ Snag saved: snagisk.example/s/8f2k 0:42 of video · 12 console logs · 31 requests · 18 actionsWhat's in a snag
Every bug report carries the evidence.
A snag is one link with what you need to reproduce the bug, whether an agent or a person opens it.
Video
The screen as it happened, lined up with everything else on the timeline.
Console logs
Every console message, with a timestamp.
Network requests
Every request and response, including the failed ones.
Actions
Each click and keystroke, written up as repro steps.
Environment
Browser, OS, viewport and URL.
For people
Capture a bug yourself.
Snags work without an agent too. Capture the bug in your browser and share the link. A teammate or an agent picks it up from there.
Checkout button does nothing
Captured by youChrome · macOS · 1440×900 · /checkout
Console
Network