Feedthrough
Let your coding agent inspect live DOM, console logs, and network traffic inside the app you are debugging in any browser.
Overview
Feedthrough MCP is an MCP server for the Ship phase that bridges your web app so agents read DOM, console, and network data over a local WebSocket.
What is this MCP server?
- npm package @feedthrough/mcp with stdio MCP transport
- WebSocket bridge reads DOM, console, and network from your running app
- Works with any browser once the in-app bridge is connected
- FEEDTHROUGH_PORT env (default 8765) for the bridge server
- FEEDTHROUGH_ALLOWED_HOST_SUFFIXES controls which hosts may connect
- npm package @feedthrough/mcp version 0.3.1
- Default bridge port 8765 via FEEDTHROUGH_PORT
Community signal: 2 GitHub stars.
What problem does it solve?
Describing front-end bugs to an agent without live runtime context wastes turns on guesswork and incomplete repro steps.
Who is it for?
Indie devs debugging React, Vue, or vanilla apps locally or on .test domains with Claude Code or Cursor at the keyboard.
Skip if: Headless CI-only pipelines with no interactive browser, or teams that cannot run an npm stdio MCP beside the app.
What do I get? / Deliverables
Your agent inspects the same DOM, console, and network events you see while you test fixes in the browser.
- Agent-readable DOM snapshots and structure from a live session
- Console error and log context in the agent thread
- Network activity visibility for reproducing API and asset failures
Recommended MCP Servers
Journey fit
Shipping web products breaks at the boundary between code and runtime; this bridge closes that gap during test and fix cycles before launch. Testing is the canonical shelf because the server exists to observe real browser behavior while you reproduce bugs and verify fixes.
How it compares
In-app debug bridge MCP, not Playwright cloud automation or a static code-review skill.
Common Questions / FAQ
Who is Feedthrough MCP for?
Solo builders shipping web apps who want agents to see live page state instead of relying on copied logs.
When should I use Feedthrough MCP?
Use it during manual test sessions when UI, console, or network bugs need tight feedback loops before release.
How do I add Feedthrough MCP to my agent?
Install @feedthrough/mcp from npm, configure stdio in your MCP client, set FEEDTHROUGH_PORT if needed, wire the bridge in your app, and allow host suffixes via FEEDTHROUGH_ALLOWED_HOST_SUFFIXES.