Browserbase Find Templates

Purpose

Discover and list the starter templates Browserbase publishes for bootstrapping browser-automation projects — return each template's slug, supported language(s) (TypeScript / Python / Go), and the canonical clone command. Read-only enumeration; no scaffolding occurs unless the caller subsequently invokes bb templates clone.

When to Use

• Answering "what Browserbase templates exist for {use case}?" (e.g. form-filling, scraping, MFA, captchas, proxies, gemini/cua agents).
• Bootstrap-time discovery: a developer asks for a starter project before writing any code.
• Cross-language coverage check: does template X exist in both Python and TypeScript? (Most do; a handful are TS-only.)
• Periodic catalog refresh — the repo's default branch (dev) is updated weekly; pulling bb templates list is the cheapest way to detect new additions.

Workflow

Browserbase exposes a first-party CLI (@browserbasehq/cli, binary bb) whose templates subcommand reads the canonical catalog directly from github.com/browserbase/templates (default branch dev). The CLI is the recommended path — output is structured, no JS rendering, no anti-bot surface, and BROWSERBASE_API_KEY is not required for templates list/clone (the CLI hits GitHub, not the Browserbase platform API, for this subcommand). The public website https://www.browserbase.com/templates is a JS-rendered marketing surface over the same repo; treat it as a fallback only when the CLI is unavailable.

1. Install the CLI (one-time, ~25 packages)

npm install -g @browserbasehq/cli
# Binary is `bb`. The system-prompt-mentioned `browse` is a separate, optional
# package (`@browserbasehq/browse-cli`) used for driving browser sessions —
# NOT required for the templates subcommand.

2. List all templates

bb templates list

Output (sectioned by language, alphabetical slugs):

typescript (40 templates):
  agent-with-human-in-loop
  amazon-global-price-comparison
  ...
  website-link-tester

python (36 templates):
  amazon-global-price-comparison
  ...
  website-link-tester

Scaffold a template with: bb templates clone <slug> --language <language>

3. Filter by language

bb templates list --language typescript    # 40 templates as of 2026-05-18
bb templates list --language python        # 36 templates as of 2026-05-18

The --language flag accepts exactly typescript or python. Go templates exist in the repo (go/hackernews) but the CLI does not expose them — see Site-Specific Gotchas.

4. (Optional) Scaffold a chosen template

Once a slug is identified, the user can clone it:

bb templates clone <slug> [--language <python|typescript>] [<destination-path>]
# Defaults: destination = ./<slug>, language = first available match
# Examples:
bb templates clone form-filling --language typescript
bb templates clone amazon-product-scraping --language python ./my-scraper

clone writes a ready-to-run project directory; the skill itself stops at listing.

5. Cross-validate against the source repo (optional integrity check)

The catalog source of truth is https://github.com/browserbase/templates on branch dev. To programmatically verify the list (e.g. detect templates added since the locally-installed CLI version was published):

curl -fsS "https://api.github.com/repos/browserbase/templates/contents/typescript?ref=dev" \
  | jq -r '.[] | select(.type=="dir") | .name'
curl -fsS "https://api.github.com/repos/browserbase/templates/contents/python?ref=dev" \
  | jq -r '.[] | select(.type=="dir") | .name'
curl -fsS "https://api.github.com/repos/browserbase/templates/contents/go?ref=dev" \
  | jq -r '.[] | select(.type=="dir") | .name'

GitHub's unauthenticated API permits 60 req/hour — sufficient for daily catalog sync.

Browser fallback

If the CLI cannot be installed (sandbox restrictions, npm offline, etc.):

1. Open https://www.browserbase.com/templates in a Browserbase remote session. A bare session is sufficient — Browserbase's own marketing site is not anti-bot-gated and a residential proxy / advanced stealth is not required.
2. The page is fully client-side rendered (Next.js); bb fetch returns the prerender HTML shell (~150 KB) but no template slugs are present in the static HTML — they are hydrated from a runtime JSON payload. You must drive a real browser, not fetch.
3. Wait for the templates grid to mount, then capture browse get markdown body and split per-card to extract template title, description, and View on GitHub href. Each card's GitHub link decodes the language: paths under /typescript/<slug> vs /python/<slug> vs /go/<slug>.
4. As a no-JS alternative: list the GitHub repo directly (Step 5 above) — that requires only curl/fetch and returns the canonical slug list without rendering.

The browser path costs ~6× more turns and surfaces no information the CLI doesn't already expose. Use only when the CLI is genuinely unavailable.

Site-Specific Gotchas

• CLI binary is bb, not browse. The npm package @browserbasehq/cli installs a single binary called bb. A separate package @browserbasehq/browse-cli installs browse for browser-driving verbs. The templates subcommand lives on bb, not browse.
• templates list does NOT require BROWSERBASE_API_KEY. Unlike bb fetch, bb search, and bb sessions ..., the templates subcommand reads from GitHub, not the Browserbase platform API. It works offline-of-Browserbase as long as api.github.com is reachable. (Other bb subcommands DO require the env var and will exit with "Missing Browserbase API key" otherwise.)
• Go templates exist in the repo but the CLI hides them. bb templates list only enumerates typescript/ and python/. The repo also contains go/hackernews (and may grow). The --language flag's accepted values are hardcoded to python | typescript; Go templates are reachable only by cloning the repo directly: git clone --depth 1 -b dev https://github.com/browserbase/templates.git.
• Default branch is dev, not main. Direct GitHub API calls must pass ?ref=dev or the response will reflect a stale main snapshot. (The repo also marks main as vanta_production_main for compliance reasons — dev is where templates are actually pushed.)
• Counts drift weekly. As of 2026-05-18: 40 TypeScript + 36 Python + 1 Go template. The repo was last pushed 2026-05-15 and is under active development (17 open issues, 12 forks). Do not hard-code counts in downstream consumers — re-run bb templates list per discovery cycle.
• https://www.browserbase.com/templates returns 200 + ~150 KB of HTML shell, but the template grid is hydrated at runtime. bb fetch <url> (the Browserbase Fetch API, an HTTP-only path) returns the shell with no template slugs in it. The slugs are only visible after JS execution. Don't waste turns on fetch-based scraping of the marketing page — use the CLI or the GitHub API.
• Template slugs do NOT always exist in both languages. Cross-language coverage as of 2026-05-18: TypeScript has agent-with-human-in-loop, browser-agent-demo, dynamic-form-filling, gemini-3-flash, microsoft-cua, puppeteer that have no Python counterpart. Python has cartesia-form-filling, cerebras-docs-checker that have no TypeScript counterpart. The remaining ~35 slugs exist in both. If a user asks for "the X template in Python" and it's TS-only, surface the asymmetry explicitly rather than failing the clone.
• CLI's --language flag is required for clone only if the slug exists in multiple languages. For single-language slugs, the CLI auto-selects. For dual-language slugs, omitting --language causes an interactive prompt — pass it explicitly in non-interactive contexts.
• No category/tag metadata in the CLI output. Slugs are surfaced as flat alphabetical lists. To classify templates by topic (e.g. "all MFA templates" → manual-mfa-with-contexts, mfa-handling, playwright-mfa-handling) you must pattern-match the slug itself or fetch each template's README from GitHub.

Expected Output

A canonical structured listing — flatten the CLI's two-section output into a single array, key by slug, attach the languages it supports.

{
  "source": "bb templates list",
  "cli_version": "0.5.7",
  "repo": "github.com/browserbase/templates",
  "branch": "dev",
  "counts": { "typescript": 40, "python": 36, "go": 1, "unique_slugs": 43 },
  "templates": [
    {
      "slug": "form-filling",
      "languages": ["typescript", "python"],
      "clone_commands": [
        "bb templates clone form-filling --language typescript",
        "bb templates clone form-filling --language python"
      ],
      "github_url": "https://github.com/browserbase/templates/tree/dev/typescript/form-filling"
    },
    {
      "slug": "microsoft-cua",
      "languages": ["typescript"],
      "clone_commands": [
        "bb templates clone microsoft-cua --language typescript"
      ],
      "github_url": "https://github.com/browserbase/templates/tree/dev/typescript/microsoft-cua"
    },
    {
      "slug": "cartesia-form-filling",
      "languages": ["python"],
      "clone_commands": [
        "bb templates clone cartesia-form-filling --language python"
      ],
      "github_url": "https://github.com/browserbase/templates/tree/dev/python/cartesia-form-filling"
    },
    {
      "slug": "hackernews",
      "languages": ["go"],
      "clone_commands": [
        "git clone --depth 1 -b dev https://github.com/browserbase/templates.git && cp -r templates/go/hackernews ./hackernews"
      ],
      "github_url": "https://github.com/browserbase/templates/tree/dev/go/hackernews",
      "note": "Not surfaced by `bb templates list`; CLI clone unsupported as of v0.5.7."
    }
  ]
}

When the caller only asks "what templates are there?" (no schema specified), a degenerate form is acceptable:

{
  "typescript": ["agent-with-human-in-loop", "amazon-global-price-comparison", "..."],
  "python":     ["amazon-global-price-comparison", "amazon-product-scraping", "..."],
  "go":         ["hackernews"]
}