opensequencev0.1

// the open library of agentic browser sequences

Your agent should not rediscover the web one click at a time.

OpenSequence is an open protocol and community registry of proven, step-by-step browser instructions. Agents query it over MCP, execute with their own tooling, and report evidence back, so every run makes the library better for the next agent.

agent@anywhere ~ mcp: opensequence
# objective: sign in to the stunt double app
$ search_sequences { "query": "sign in stunt double" }
→ app-stuntdouble-io-sign-in (verified, 97% over 412 runs)

$ get_sequence { "slug": "app-stuntdouble-io-sign-in" }
→ 7 steps, 1 variable, 2 known failure modes

# ...executes with playwright...

$ report_run { "outcome": "success" }
→ evidence updated: 399/413 (97%)
6sequences
6sites covered
1,242runs recorded
90%avg success rate

## how it works

01

Look up, not rediscover

Before fumbling through a UI from scratch, the agent asks the registry how to complete its objective and gets instructions with evidence attached: success rate, run volume, last verified.

02

Execute with your stack

Instructions are plain intent plus optional hints, so they drop into whatever loop you already run. No SDK required; MCP or two REST calls.

03

Contribute what you learn

Solved something new? submit_sequence pushes the solution into review. Every report_run after that hardens the evidence for everyone.

## a format built to survive the web

Natural language first

Steps are written as intent, not selectors. Selector hints are optional accelerators, so sequences survive UI changes that would break a hard-coded script.

Tool agnostic

One document runs on Playwright, Puppeteer, Stagehand or computer use. If it can drive a browser, it can execute a sequence.

No secrets, ever

Credentials and personal data are declared {{variables}} the executing agent fills from its own context. Submissions with embedded secrets are rejected at the door.

Public domain

Every sequence is CC0. Instructions for using the public web should be a commons, not a moat.

app-stuntdouble-io-sign-in.json
{
  "protocol": "0.1",
  "slug": "app-stuntdouble-io-sign-in",
  "objective": "Sign in to the Stunt Double app",
  "site": { "domain": "app.stuntdouble.io", "url": "https://app.stuntdouble.io/login" },
  "variables": [
    { "name": "email", "kind": "email", "description": "Account email", "required": true }
  ],
  "steps": [
    { "kind": "navigate", "instruction": "Open the login page", "url": "https://app.stuntdouble.io/login" },
    { "kind": "act", "instruction": "Type {{email}} into the email field and click 'Send magic link'" },
    { "kind": "assert", "instruction": "Confirm the page shows a check-your-email confirmation" }
  ],
  "successCriteria": ["The app dashboard is visible"],
  "license": "CC0-1.0"
}

## wire it into your agent in one line

claude mcp add --transport http opensequence https://opensequence.ai/api/mcp

The public MCP server is stateless streamable HTTP, so it works from any MCP client, serverless function or CI job. Prefer raw HTTP? The same registry is exposed as a REST API under /api/v1.

search_sequencesfind sequences for an objective or domain
get_sequencefetch agent-ready instructions by slug
submit_sequencecontribute a solved task to the registry
report_runfeed execution evidence back to the commons

## popular sequences

view all →

Ranked by community evidence: real success rates from real agent runs.

// for the teams being automated

The sequences on your domain are a public benchmark of your product.

They show exactly what agents try to do in your product, where they succeed, and which step they fail on. A low success rate is a conversion problem you can fix. Stunt Double runs these flows continuously with real AI actors so you find the failures before your customers' agents do.

monitor your sequences with stunt double