Skip to content

harness-feGive your AI agent eyes, hands — and your source map

A dev-time harness that lets an MCP agent see your frontend's console, network, and DOM, drive the page, and trace every element back to the exact file and line. Report → fix → verify, in one loop.

An AI agent (Claude · Codex · Cursor) at the center, running an autonomous loop — harness-fe feeds back what it sees and accepts the agent's control

Built for the dev-time agent loop

harness-fe isn't a production monitor and isn't a general-purpose browser bot. It's the missing layer that turns an AI coding agent into one that can see what your app is doing, drive it, and know exactly which source line to fix — then prove the fix by replaying the flow. Drop in one build plugin, point an MCP-aware agent at it, and the report → fix → verify loop closes itself.

How it compares

Other tools each cover a slice. harness-fe is the one built end-to-end for the developer's agent loop — source-aware, full-stack, and dev-only.

harness-feChrome DevTools MCPbrowser-use & co.Sentry / LogRocket
Built fordev-time agent loopbrowser debuggingend-user task agentsproduction monitoring
Source-aware (file : line)
Report → fix → verify loop
Full-stack observability¹
Drive the page (agent)
Session replay (rrweb)
Multi-bundler / framework²
MCP-native
Dev-only · zero prod footprint
Team isolation + governance

yes  ·  partial / conditional  ·  no  ·  n/a   |   ¹ console + network + WebSocket + errors + DOM recordings  ·  ² Vite · Webpack · Rspack · Next.js · Vue · React

Three minutes to your first agent-driven session → Quickstart · Team mode · Migrating from 3.x

Released under the MIT License.