proving ground

Agent playground

This page is a course an agent can run, in your browser: discover the site-wide WebMCP catalog, call the in-page window.__tools API, and return the structured results those tools promise. The basics are confirmed on load; three challenges are left for your agent.

This page as markdown:/for-agents/playground.md.

Environment

webmcp: enable JavaScript to probe
window.__tools: enable JavaScript to probe

The page probes document.modelContext (the canonical WebMCP surface) and falls back to navigator.modelContext (older preview runtimes). Where neither exists the check is skipped, not failed — the window.__tools path works in any browser.

WebMCP catalog

The catalog is the site-wide WebMCP toolset; slugs become snake_case tool names.

tool	page	does
base64	/tools/base64	Unicode-safe Base64 encode and decode, with URL-safe variant.
url	/tools/url	Percent-encode and decode URLs and query components.
jwt	/tools/jwt	Decode a JWT's header, payload and time claims — or encode and HMAC-sign a new token. All in the browser.
hash	/tools/hash	SHA-1 / SHA-256 / SHA-384 / SHA-512 of text or files, via WebCrypto.
radix	/tools/radix	Binary, octal, decimal, hex — arbitrary precision via BigInt.
timestamp	/tools/timestamp	Epoch seconds/millis ⇄ human dates, timezone-aware, with relative time.
cron	/tools/cron	Cron expression → plain-English schedule + the next run times.
uuid	/tools/uuid	UUID v4 and v7, single or bulk, copy-ready.
json	/tools/json	Pretty-print, minify, validate — with precise error positions.
case	/tools/case	camelCase, snake_case, kebab-case, Title Case, CONSTANT_CASE and more.
diff	/tools/diff	Line and word diff of two texts, computed locally.
regex	/tools/regex	Live match highlighting, capture groups, and a common-pattern library.
markdown	/tools/markdown	Split-pane markdown editor — live preview, right-to-left aware, copy as HTML, Markdown, or rich text.
count	/tools/count	Words, characters, lines, sentences, reading time — plus a live token count (GPT o200k by default, switchable to cl100k or Llama 3).
plot	/tools/plot	Plot functions on an interactive canvas — pan, zoom, multiple expressions, shareable links.
lorem	/tools/lorem	Filler text by the word, sentence, or paragraph — copy-ready.
csv	/tools/csv	CSV to JSON and back — quoted fields, custom delimiters, header-row toggle. Values stay strings.
color	/tools/color	Hex, rgb, hsl, and oklch — plus a tints-and-shades palette and a WCAG contrast checker.
data	/tools/data	Convert between JSON, YAML, TOML, and TOON — auto-detects the input, points at the exact parse error, and shows the token-count savings of TOON's compact tabular arrays.
qr	/tools/qr	Text or URL → QR code as crisp SVG, with download. Decodes images too, where the browser can.
contact	/tools/contact	Compose a prefilled email draft, plus quick contact links — email, GitHub, LinkedIn. Opens in your own mail client.

21 tools at build · re-enumerated live by the checks below

Scorecard

scorecard idle — checks run when JavaScript loads

automated checks

What this page confirms about your browser's agent layer — checked on load, again on demand (run({ action: "checks" }) or the button). Local and deterministic, each compared to a known answer:

WebMCP runtime presentpending
queued — runs on load
tool catalog registered site-widepending
queued — runs on load
window.__tools API live on this pagepending
queued — runs on load
describe() honors the contract shapepending
queued — runs on load
deterministic run matches the pinned valuepending
queued — runs on load
async run resolves the pinned digestpending
queued — runs on load
run() output carries its published keyspending
queued — runs on load

agent challenges

Three challenges the page cannot do for you — each exercises a real invocation path, and the playground grades the submission. Every one runs on the same challenge phrase:

agent proving ground

agent-describecall describe() on the base64 toolpending
Read the base64 tool's contract — window.__tools["base64"].describe() on /tools/base64, or fetch /tools/base64.json — and submit the array of its params keys.
awaiting your submission
agent-base64run the base64 tool on the phrasepending
Encode the challenge phrase with the base64 tool — the site-wide WebMCP base64 tool works from this very page, or window.__tools["base64"].run({ input }) on /tools/base64, or the URL form /tools/base64#input=… — and submit the encoded output.
awaiting your submission
agent-hashhash the phrase with SHA-256pending
Digest the challenge phrase with the hash tool — the WebMCP hash tool, or window.__tools["hash"].run({ input, algo: "SHA-256" }) on /tools/hash (async — await it) — and submit the hex digest.
awaiting your submission

Submit with run({ action: "submit", challenge: "<id>", result: <your value> }). The scorecard row flips live; a tool's structured return is accepted as well as the bare string, and re-submission is allowed.

Drive it headlessly

The scorecard is driveable headlessly. The page registers window.__tools["playground"] with the standard run()/describe() shape — run() returns structured values and is async, so await it:

const pg = window.__tools["playground"];
pg.describe(); // the contract for this API
await pg.run({ action: "checks" }); // re-run the automated suite
const { phrase, challenges } = await pg.run({ action: "challenges" });
// … exercise a tool on the phrase, then:
await pg.run({ action: "submit", challenge: "agent-base64", result: "…" });

The contract this page proves lives at/for-agents. Tool contracts:/SKILL.mdand thetools manifest.