@agentutility/mcp-browser-workflow

@agentutility/mcp-browser-workflow

Workflow memory for brittle browser and desktop agents.

Frontier probe for turning screenshots, Playwright traces, DOM snapshots, and recorded task paths into reusable workflow hints, form-fill plans, change diffs, and failure evidence.

Pricing: pay-per-call in USDC on Base. No subscriptions, no API keys. See per-tool prices below.

Install — Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "agentutility-browser-workflow": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-browser-workflow"],
      "env": { "X402_PRIVATE_KEY": "0xYOUR_PRIVATE_KEY_HEX" }
    }
  }
}

Restart Claude Desktop. 3 tools appear in the tool palette.

Install — Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "agentutility-browser-workflow": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-browser-workflow"],
      "env": { "X402_PRIVATE_KEY": "0x..." }
    }
  }
}

Funding

Send any amount of USDC on Base mainnet to the address derived from your X402_PRIVATE_KEY. The MCP server uses it to pay for tool calls automatically.

USDC on Base contract: 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913

Tools (3)

Tool	Description
browser-flow-digest	(0.02 USDC/call) Browser flow digest API that compresses a recorded trajectory or trace into reusable workflow memory: an ordered step list, the preconditions each step assumes, and failure hints for anything that broke or looked fragile. Built for trajectory-to-steps compression so an agent doesn't replay a full session log every time it repeats a task. Send a trajectory (JSON array of recorded browser actions, or a plain-text trace) and an optional goal, and get back a named workflow with steps, preconditions, failure hints, and reusability notes grounded only in what the trajectory actually shows. Use it as a workflow memory API, a browser automation trace summarizer, or a reusable-recipe generator for repeat browser agent tasks.
dom-change-diff	(0.01 USDC/call) DOM diff for two HTML snapshots of the same page or component, built for browser-automation scripts whose CSS selectors quietly break after a site redesign. Send before_html and after_html plus the selectors your workflow watches, and get back per-selector added/removed/moved/attribute-changed status, stale DOM detection for targets that vanished, and selector stability hints: replacement selectors built from an element's id, data-testid, or aria-label so a broken automation step can self-repair. Use it as a browser workflow repair tool, a selector-stability checker, or a pre-flight check before re-running a scraping or RPA script against a site that may have changed.
form-fill-plan	(0.02 USDC/call) Form fill plan API that turns an HTML form plus a plain-language goal into an ordered browser form automation plan: selector, field label, value, and action (fill/select/check/uncheck/click) for each step. Built for autofill steps and Playwright form plan generation ahead of a real run. Send form_html, a goal like 'sign up for the newsletter with my email', and any known data values, and get back a step-by-step plan with unresolved fields flagged rather than guessed. Advisory only: the plan is heuristic guidance from static HTML, not a guarantee, and should be verified against the live DOM before executing. Use it as a browser form automation API, an autofill planning tool, or a pre-run checklist generator for RPA and agent browser workflows.

How it works

• Agent calls a tool (e.g. browser-flow-digest).
• MCP server POSTs to https://x402.agentutility.ai/browser-flow-digest.
• The endpoint responds HTTP 402 with payment instructions.
• The MCP server signs an EIP-3009 USDC transfer authorization with X402_PRIVATE_KEY and retries.
• CDP facilitator settles on Base.
• The endpoint returns the actual response.

The agent never sees the payment flow — it just gets the result.

Links

• Cluster overview: https://agentutility.ai/browser-workflow/
• All MCP packages: https://mcp.agentutility.ai/
• Source: https://github.com/rooz21/x402/tree/main/packages/mcp-browser-workflow

Version: 0.1.0 · License: MIT