Scout
Browser automation, one binary. The simpler alternative to Playwright — no Node, no Python, no runtime. Drive a real browser from Go, any shell, any AI agent (built-in MCP server), or a chat UI.
A single statically-linked `scout` binary gives you a CLI, an 87-tool MCP server (so any MCP-aware agent — Claude Desktop, Cursor, Cline, custom — has a browser), a conversational chat UI, and a Go library with Gin-like middleware composition. Same engine, four access points.
```bash
brew install --cask klarlabs-studio/tap/scout-mcp
```
## vs. Playwright
| | Scout | Playwright |
|---|---|---|
| Install | one ~15 MB binary | npm + ~600 MB browser cache |
| Runtime dep | **none** (static) | Node.js always; Python/Java/.NET as wrappers |
| Drive from | Go, any shell, MCP, chat UI | TS/JS first-class; others lag |
| AI-agent native | **built-in** `scout mcp serve` | separate `playwright-mcp` project |
| Token-aware extraction | DOM diff, distillation, observation budgets (50–80% fewer tokens) | not provided |
| Action playbooks | record & replay deterministic JSON | codegen produces a script you maintain |
| Container deploy | drop into `scratch` or `distroless` | carry Node + browser binaries |
| CDP access | direct WebSocket, zero abstraction | internal protocol over CDP |
## Quick Start
```bash
# CLI — visible browser, one-shot commands
scout observe https://example.com # structured page snapshot
scout markdown https://news.ycombinator.com # page as compact markdown
scout screenshot https://github.com # save screenshot
scout extract https://example.com h1 # extract element text
scout frameworks https://react.dev # detect React, Vue, etc.
# MCP Server — give AI agents browser superpowers
claude mcp add scout -- scout mcp serve
# Browser UI — conversational browser automation
scout ui serve --provider=ollama --model=mistral
cd ui && npm install && npm run dev # open http://localhost:3000
```
## Install
```bash
# Homebrew
brew install --cask klarlabs-studio/tap/scout-mcp
# Direct binary
curl -fsSL https://raw.githubusercontent.com/klarlabs-studio/scout/main/install.sh | bash
# Go
go install go.klarlabs.de/scout/cmd/scout@latest
# As a library
go get go.klarlabs.de/scout
```
## MCP Server — 77 Tools
Run `scout mcp serve` and any MCP-aware agent has a browser. No second project to install, no Node runtime, no Python interpreter — the binary is the server. Configure in any MCP client:
```bash
claude mcp add scout -- scout mcp serve # Claude Code
```
```json
{"mcpServers": {"scout": {"command": "scout", "args": ["mcp", "serve"]}}}
```
### Tool Categories
| Category | Tools |
|----------|-------|
| **Navigation** | `navigate`, `observe`, `observe_diff`, `observe_with_budget` |
| **Interaction** | `click`, `click_label`, `click_text`, `type`, `hover`, `double_click`, `right_click`, `select_option`, `scroll_to`, `scroll_by`, `focus`, `drag_drop`, `dispatch_event` |
| **Forms** | `fill_form`, `fill_form_semantic` (checkbox/radio + state echo), `discover_form` |
| **Extraction** | `extract`, `extract_all`, `extract_table`, `auto_extract`, `scroll_and_collect`, `markdown`, `readable_text`, `accessibility_tree` |
| **Capture** | `screenshot`, `annotated_screenshot`, `pdf` |
| **Network** | `enable_network_capture`, `network_requests` |
| **Tabs** | `open_tab`, `switch_tab`, `close_tab`, `list_tabs` |
| **Frameworks** | `wait_spa`, `detect_frameworks`, `component_state`, `app_state` |
| **Playback** | `start_recording`, `stop_recording`, `save_playbook`, `replay_playbook` |
| **Video** | `start_screen_recording`, `stop_screen_recording` |
| **Smart Helpers** | `check_readiness`, `suggest_selectors`, `session_history` |
| **Vision** | `hybrid_observe`, `find_by_coordinates` |
| **Batch** | `execute_batch` |
| **Iframe** | `switch_to_frame`, `switch_to_main_frame` |
| **Trace** | `start_trace`, `stop_trace` |
| **Cookies** | `cookies_list`, `cookies_clear`, `cookies_set`, `dismiss_cookies` |
| **Diagnostics** | `detect_dialog`, `detect_auth_wall`, `console_errors` (incl. network 4xx/5xx), `failed_requests`, `compare_tabs`, `upload_file` |
| **Utility** | `has_element`, `wait_for`, `configure`, `set_viewport`, `web_vitals`, `select_by_prompt` |
All tools have MCP annotations (`ReadOnly`, `OpenWorld`, `ClosedWorld`, `Idempotent`) for smart auto-approval. Read-only tools like `observe`, `extract`, and `screenshot` run without permission prompts.
### Runtime Configuration
Switch between headless and visible browser without restarting, and opt into local-dev workflows (loopback, private IPs):
```
Agent: configure(headless: false) → browser window appears
Agent: navigate("https://...") → watch it work
Agent: configure(headless: true) → back to headless
Agent: configure(allow_private_ips: true) → unlock localhost / 192.168.* / 10.*
Agent: navigate("http://127.0.0.1:4173/") → drive your local dev server
```
The MCP server also reads `SCOUT_ALLOW_PRIVATE_IPS=1` at startup as a one-shot toggle for trusted environments.
### Screen Recording (video)
Record the active page as a video. Pure CDP — works in headless, no Playwright needed. Recording survives `navigate`, `open_tab`, and `switch_tab` calls in between, so a multi-page demo lands as one continuous clip:
```
Agent: start_screen_recording({ width: 1280, height: 800, fps: 15, format: "webm" })
Agent: navigate("https://example.com")
Agent: click("#cta")
Agent: navigate("https://example.com/dashboard") # recording continues across pages
Agent: stop_screen_recording()
→ { path: "/tmp/scout-rec-XXX.webm", format: "webm", encoder: "ffmpeg",
frame_count: N, duration_ms: N }
```
If `ffmpeg` is on PATH, the result is encoded to WebM (libvpx-vp9) or MP4 (libx264). If not, scout returns the raw JPEG frames directory plus an ffmpeg concat list so you can encode offline. The result is always a file path, never base64 — never enters your LLM token budget.
Realistic FPS: ~10–15 on typical pages, capped at 30. Implementation polls `Page.captureScreenshot` (CDP `Page.startScreencast` events are silently dropped under `--headless=new` Chrome).
## Browser UI
A conversational browser automation interface. Type natural language, watch the browser respond in real-time.
```bash
# Start the AG-UI server (Go backend)
scout ui serve --provider=ollama --model=mistral # local, no API key
scout ui serve --provider=claude # needs ANTHROPIC_API_KEY
scout ui serve --provider=openai --model=gpt-4o # needs OPENAI_API_KEY
scout ui serve --provider=groq --base-url=https://api.groq.com/openai --model=llama-3.3-70b-versatile
# Start the Vue frontend
cd ui && npm install && npm run dev # http://localhost:3000
```
The UI streams AG-UI protocol events over SSE:
- **Chat panel** with markdown rendering and quick-action pills
- **Live browser viewport** with screenshot streaming and URL bar
- **Activity timeline** showing tool calls in real-time
- **Stop button** to cancel mid-stream
The Go server handles the agentic loop: LLM decides which scout tools to call, executes them, streams browser state deltas back to the frontend. Supports any OpenAI-compatible endpoint via `--base-url`.
## Agent Package (Go)
High-level Go API for callers that want to embed scout in a program. Structured output, auto-wait, goroutine-safe. Most users reach scout through the CLI or MCP server above — this section is for the Go-library path.
```go
session, _ := agent.NewSession(agent.SessionConfig{Headless: true})
defer session.Close()
// Navigate and observe
session.Navigate("https://example.com")
obs, _ := session.Observe() // links, inputs, buttons, text + action costs
// DOM diff — only what changed (saves 50-80% tokens)
session.Click("#submit")
_, diff, _ := session.ObserveDiff()
// diff.Classification: "modal_appeared"
// diff.Summary: "Modal/dialog appeared: Login required"
// Semantic form filling — no CSS selectors
session.FillFormSemantic(map[string]string{
"Email": "user-example", "Password": "secret",
})
// Visual grounding — click by number, not selector
result, _ := session.AnnotatedScreenshot() // numbered labels on elements
session.ClickLabel(7) // click element [7]
// Multi-tab coordination
session.OpenTab("pricing", "https://example.com/pricing")
session.SwitchTab("default")
// Framework detection (19 frameworks)
frameworks, _ := session.DetectedFrameworks() // ["react", "nextjs"]
state, _ := session.ComponentState("#app") // read React/Vue state
// Network capture — read API responses directly
session.EnableNetworkCapture("/api/")
captured := session.CapturedRequests("/api/users")
// Action replay — record once, replay without LLM
session.StartRecordingPlaybook("login-flow")
// ... do stuff ...
pb, _ := session.StopRecordingPlaybook()
agent.SavePlaybook(pb, "login.json")
// Later: session.ReplayPlaybook(pb) // 100x cheaper
// Persistent profiles
session.SaveProfile("session.json") // cookies + localStorage
session.LoadProfile("session.json")
// Content distillation (5 levels)
session.Markdown() // ~2-8KB compact markdown
session.ReadableText() // ~1-4KB main content only
session.AccessibilityTree() // ~1-4KB semantic tree
session.ObserveWithBudget(500) // fit in ~500 tokens
```
## Core Library (Go)
Gin-like Engine/Context/Group/HandlerFunc with middleware composition. The lowest-level Go API — use it when you want full control of task lifecycle, named groups, and middleware chains:
```go
engine := browse.Default(browse.WithHeadless(true))
engine.MustLaunch()
defer engine.Close()
engine.Use(middleware.Stealth())
engine.Use(middleware.Retry(middleware.RetryConfig{MaxAttempts: 3}))
engine.Use(middleware.Timeout(30 * time.Second))
admin := engine.Group("admin", middleware.BasicAuth("admin", "secret"))
admin.Task("export", func(c *browse.Context) {
c.MustNavigate("https://app.example.com/admin")
table, _ := c.ExtractTable("#users")
c.Set("data", table)
})
engine.RunGroup("admin")
```
### Middleware
| Category | Middleware |
|----------|-----------|
| **Resilience** | `Retry`, `Timeout`, `CircuitBreaker`, `RateLimit`, `Bulkhead` |
| **Auth** | `BearerAuth`, `BasicAuth`, `CookieAuth`, `HeaderAuth` |
| **Anti-detection** | `Stealth` (10 patches: webdriver, plugins, WebGL, etc.) |
| **Network** | `BlockResources`, `WaitNetworkIdle` |
| **Utilities** | `ScreenshotOnError`, `SlowMotion`, `Viewport` |
## CLI
The CLI is a deliberate **one-shot tier**: each command launches a browser,
navigates, runs one read/diagnostic action, prints the result, and exits.
Stateful interactive flows (click→type→submit, multi-tab, live cookie/network
manipulation) live in the MCP server (`scout mcp serve`) and chat UI
(`scout ui serve`), which keep a session alive across actions.
CLI defaults to visible browser (`--headless` to hide):
```bash
scout navigate # page info as JSON
scout observe # structured observation
scout markdown # compact markdown
scout readable # main readable content (boilerplate stripped)
scout accessibility # semantic accessibility tree
scout screenshot [--output f] # save screenshot
scout pdf [--output f] # save PDF
scout extract # extract text
scout table # extract a table as structured rows
scout auto-extract # auto-detect dominant data pattern
scout eval # run JavaScript
scout form discover # discover form fields
scout frameworks # detect frameworks
scout app-state # global app/store state (JSON)
scout aria # ARIA violation report (JSON)
scout vitals # Core Web Vitals (JSON)
scout console # console errors + network 4xx/5xx (JSON)
scout dialog # detect visible modal/dialog (JSON)
scout auth-wall # detect login/auth wall (JSON)
scout cookies # list cookies (values redacted)
scout watch [--interval=5s] # live-watch page changes
scout pipe [selector] # batch process URLs from stdin
scout record [--output f] # interactive recording → playbook
scout mcp serve # start MCP server
scout ui serve [flags] # start chat UI
scout version # print version
```
## Architecture
```
scout/
├── browse.go, engine.go, context.go # Gin-like API
├── page.go, selection.go # CDP page & element interaction
├── recorder.go # Action playbook recording (Navigate/Click/Type → JSON)
├── middleware/ # stealth, resilience, auth, network
├── agent/ # AI agent API (50+ methods)
│ ├── session.go # Session lifecycle, Navigate, Click, Type
│ ├── observe.go, diff.go # Observe, ObserveDiff, cost estimation
│ ├── content.go # Markdown, ReadableText, AccessibilityTree
│ ├── form.go # DiscoverForm, FillFormSemantic, MatchFormField
│ ├── annotate.go # AnnotatedScreenshot, ClickLabel
│ ├── network.go # EnableNetworkCapture, CapturedRequests
│ ├── spa.go # DetectedFrameworks, ComponentState, GetAppState
│ ├── tabs.go # OpenTab, SwitchTab, CloseTab, ListTabs
│ ├── playbook.go # StartRecording, ReplayPlaybook, SavePlaybook
│ ├── interact.go # Hover, DragDrop, SelectOption, ScrollTo
│ ├── profile.go # CaptureProfile, ApplyProfile, SaveProfile
│ ├── selector.go # Playwright :text() selector translation
│ ├── budget.go # ObserveWithBudget, EstimateTokens
│ ├── nlselect.go # SelectByPrompt, fuzzy NL element matching
│ ├── batch.go # ExecuteBatch, sequential multi-action
│ ├── vision.go # HybridObserve, FindByCoordinates
│ ├── trace.go # StartTrace, StopTrace, action tracing
│ ├── screencast.go # StartScreenRecording / StopScreenRecording — video via captureScreenshot polling + ffmpeg encode
│ ├── iframe.go # SwitchToFrame, SwitchToMainFrame
│ └── vitals.go # WebVitals (LCP/CLS/INP)
├── internal/cdp/ # WebSocket CDP client (context-aware)
├── internal/launcher/ # Chrome process management
├── cmd/scout/ # CLI + MCP server (87 tools)
└── docs/ # Landing page (GitHub Pages)
```
## Security
Vulnerability scanning runs on every push and PR via [`nox`](https://github.com/nox-hq/nox). Findings are uploaded to GitHub code scanning, annotated inline on PRs, and gated against `.nox/baseline.json` so regressions block merges. The status badge in the header reflects the latest main-branch scan.
`nox` also drives dependency remediation in place of Dependabot — the [Nox Remediate](.github/workflows/nox-remediate.yml) workflow runs weekly (Monday 06:00 UTC) and on demand, executing `nox fix` against fresh OSV.dev findings and opening a single PR with the verified upgrades.
```bash
# Local scan
nox scan -severity-threshold high .
# Local fix
nox fix -input findings.json
```
## License
MIT