prooftrail

ProofTrail for Coding Agents and Agent Ecosystems

This page is for one very specific search intent:

how do I give Codex, Claude Code, OpenHands, OpenCode, OpenClaw, or another coding agent a browser automation layer without losing evidence, recovery, and operator review?

ProofTrail is not trying to replace your coding agent shell.

It gives that shell a governed browser execution layer with:

• one canonical run path
• retained evidence bundles
• recovery and review surfaces
• API and MCP integration roads

Use this page when

You already use one of these patterns:

• Codex
• Claude Code
• OpenHands
• OpenCode
• OpenClaw
• another coding agent shell that can call tools or orchestrate repo tasks

And now you need browser automation that stays:

• inspectable
• replayable
• recoverable
• reviewable by humans

The short answer

Use ProofTrail when your coding agent still needs a browser lane, but you do not want that browser lane to disappear into one-shot logs or generic bot behavior.

Think of it like adding a flight recorder and recovery checklist to a vehicle:

• your coding agent still drives the broader workflow
• ProofTrail records what happened in the browser lane
• the retained run becomes something you can inspect, compare, and hand off

Truthful ecosystem fit matrix

ProofTrail does not claim first-party Codex-native, Claude-Code-native, OpenHands-native, OpenCode-native, or OpenClaw-native product integrations.

It does claim something more grounded:

Ecosystem	Most truthful fit	Best first road	What this page is not claiming
Claude Code	governed browser-evidence side road for a tool-using coding shell	MCP for Browser Automation	official vendor adapter
Codex	browser-evidence substrate with direct API control and optional MCP	API Builder Quickstart	official Codex plugin
OpenHands	browser-evidence subsystem behind a larger orchestration runtime	API Builder Quickstart	ProofTrail replacing the outer runtime
OpenCode	governed MCP browser surface behind the coding-agent shell	MCP for Browser Automation	official OpenCode adapter
OpenClaw	browser workflow backend behind a multi-channel gateway or tool router	API Builder Quickstart	first-party OpenClaw plugin

If you need a generic AI shell or a vendor-owned integration, this repo is not pretending to be that.

What these ecosystems actually get here

Across all five ecosystems, the grounded promise stays the same:

1. A canonical browser execution road
   • start from just run
   • keep one deterministic baseline instead of many unofficial scripts
2. Retained browser evidence
   • inspect manifests, summaries, and proof reports after the run
3. Recovery and review surfaces
   • explain, share, compare, and review before widening handoff
4. Two integration roads
   • API for exact contract control
   • MCP for governed tool consumption by an external AI client

Where this fits in a coding-agent stack

Layer	What stays outside ProofTrail	What ProofTrail contributes
Coding agent shell	planning, code edits, issue flow, repo reasoning	browser execution substrate
Browser lane	raw scripts, helper-only replay, brittle one-offs	canonical run plus retained evidence
Failure handling	ad-hoc debugging in logs	recovery, compare, and review packet surfaces
Tool integration	custom glue for every browser action	API and MCP surfaces over the same governed substrate

That means the most honest category fit is:

ProofTrail is browser automation for coding agents, not a coding agent replacement.

When API is the better road

Use the API road when your outer agent needs:

• exact request/response control
• contract-level integration
• direct access to runs, evidence, or automation endpoints

Start here:

1. API Builder Quickstart
2. Universal API Reference
3. node --import tsx contracts/scripts/generate-client.ts --verify

When MCP is the better road

Use the MCP road when your outer agent shell should consume tools instead of raw REST semantics.

This is the most natural fit when your coding-agent environment already thinks in tools, tasks, and delegable actions.

Start here:

1. MCP for Browser Automation
2. ProofTrail MCP Server README
3. pnpm mcp:check

Why this matters for named agent ecosystems

Coding agents and agent gateways are strong at planning, orchestration, or tool routing.

What they often still need from a browser layer is:

• proof after the run
• a stable recovery ladder
• a way to compare and explain what changed
• a governed integration surface instead of ad-hoc scripts

That is exactly where ProofTrail fits, whether the outer shell is Codex, Claude Code, OpenHands, OpenCode, OpenClaw, or another tool-using agent stack.

Honest boundary

This page does not claim:

• a first-party Codex plugin
• a first-party Claude Code plugin
• a first-party OpenHands integration
• a first-party OpenCode integration
• a first-party OpenClaw integration
• hosted collaboration for agent teams
• open-ended browser autonomy as the product default

It only claims what the repo already supports:

• evidence-first browser automation
• retained browser evidence
• recovery and review surfaces
• API and MCP integration roads
• a bounded AI reconstruction and autonomy side-road story

Suggested reading order

1. README.md
2. ProofTrail for AI Agents
3. this page
4. MCP for Browser Automation
5. Universal API Reference
6. Evidence, Recovery, and Review Workspace

This site is open source. Improve this page.