Claude Code
Give it the URL in the prompt; it fetches web pages on request. Put the URL and a one-line note in CLAUDE.md so every session knows where the reference is (Claude Code bridge guide).
The three files
| File | What it contains | Use it for |
|---|---|---|
| llms.txt | A short index: product summary, canonical URLs for the iOS, Android and macOS docs, the Cloud Builder, the vibe-coding guides and the readiness checks, plus usage notes for agents. | Orientation and a quick answer to "what does WebViewGold do"; small enough to paste into any prompt. |
| llms-full.txt | The complete documentation as plain text in one file: every iOS, Android and macOS page, every Config.swift and Config.java option, every URL scheme and injected variable, with headings like "Biometric Authentication API (iOS)". | Implementation. The agent reads the section for the feature and copies the exact scheme, parameter and variable names. |
| webmcp.json | A discovery manifest: name, description, links to llms.txt and the sitemap, a list of GET "tools" (product overview, docs hub, sitemap, vibe-coding guides) and crawl guidance. | Research agents and crawlers that need to know where the canonical pages are before reading them. |
Why agents hallucinate bridge calls
The WebViewGold bridge is unusual: you trigger native features by navigating to a URL scheme (window.location.href = "bioauth://"), and results come back as injected global variables or callback functions. Most training data describes message-handler bridges, so an agent that has not read the docs will write window.webkit.messageHandlers.share.postMessage(...) or WebViewGold.share(), both of which do nothing. The names also matter: the share sheet is shareapp://message?=TEXT&url=URL, not share://; the scanner result is barcodescanresult, not scanResult.
Giving the agent llms-full.txt removes the guessing. The file is plain text, so every agent can read it with its fetch tool, and you can verify its output with grep.
How each agent reads the docs
OpenAI Codex
The CLI and the IDE extension can download the file with a shell command when network access is allowed; cloud tasks run in a sandbox whose internet access you control, so the safest route is a committed copy under docs/vendor/ referenced from AGENTS.md.
Cursor
Add the URL as a docs source (@Docs) once; afterwards the agent can search it. @Web works for one-off questions. Mention the file in .cursor/rules so agent mode reads it before using any scheme.
Windsurf
Paste the URL or the relevant section into the conversation, or commit a local copy and reference it from the workspace rules file. Ask the agent (Devin Local, formerly Cascade) to quote the section it relied on.
Gemini CLI and Google AI Studio
Gemini CLI fetches URLs on request and reads a GEMINI.md context file; AI Studio apps can be given the section as context. Same rule: quote before you code.
My iOS WebView is finally approved. Thanks!!
A workflow that works with any agent
Vendor the documentation
Download llms-full.txt into
docs/vendor/webviewgold-llms-full.txtand commit it with the date in the commit message. A local file is faster, deterministic and available in sandboxes without network access. Refresh it when you update WebViewGold.Point the instructions file at it
One line in AGENTS.md, CLAUDE.md, .cursor/rules or the equivalent: "Before using any WebViewGold feature, read the matching section in docs/vendor/webviewgold-llms-full.txt and quote the scheme."
Prompt with a quoting requirement
Use the prompt below. Forcing the agent to quote the section first produces correct scheme names and makes a wrong guess visible before it becomes code.
Verify with grep
Run the check in the code block below: every scheme the agent used must appear in the documentation. Unknown schemes are hallucinations.
Test on a device build
Schemes do nothing in a desktop browser. Build with Xcode, Android Studio or the Cloud Builder and test on a phone; TestFlight and internal testing tracks make this quick.
A prompt that references the docs
Paste into Claude Code, Codex, Cursor, Windsurf or Gemini CLI
Before writing any code, read https://www.webviewgold.com/llms-full.txt (or docs/vendor/webviewgold-llms-full.txt if it exists) and find the section for [FEATURE, e.g. Biometric Authentication API]. Quote the exact URL scheme, parameters and the callback or variable names from that section. Then implement [FEATURE] in [FILE OR ROUTE], guarded by isNativeApp() (user agent marker MYAPPNATIVE), with a web fallback. Variables injected by the app arrive asynchronously, so poll with a timeout. If the documentation does not cover something you need, say so instead of guessing. End with the documentation headings you relied on and what I must test on a device.
Verify agent output against the docs
Run this in the repository after the agent finishes. It lists every custom scheme in your source and checks it against the documentation.
bash
# Fetch the full documentation once (or use the committed copy)
curl -sL https://www.webviewgold.com/llms-full.txt -o /tmp/wvg-docs.txt
# Every "something://" the agent used in src must exist in the docs
grep -rhoE "[a-zA-Z-]+://" src | sort -u | while read scheme; do
if grep -qF "$scheme" /tmp/wvg-docs.txt; then
echo "ok $scheme"
else
echo "UNKNOWN $scheme <- not in the WebViewGold docs"
fi
done
# Read the section for one feature before you test on a device
grep -n -A 40 "Biometric Authentication API (Face ID / Touch ID) (iOS)" /tmp/wvg-docs.txtMistakes we see in agent output
- Invented bridges:
window.WebViewGold,postMessagehandlers, Capacitor or Cordova plugins. The bridge is URL navigation. - Reading injected variables synchronously on the next line. They arrive after the navigation; poll for a few hundred milliseconds.
- Calling schemes in the browser. Guard with the user agent marker; a scheme in Safari or Chrome is a broken link.
- Using the Web Share API or a PWA push setup instead of the native share sheet and the WebViewGold push integration.
- Treating platform-specific features as universal:
settingsapp://andreconnect://are documented for Android; the biometric callbacks are documented for iOS. The file says which platform each section covers. - Hard-coding RevenueCat product IDs in several places instead of one config map, and granting entitlements on the client instead of from the webhook.
Frequently asked questions
Is there a WebViewGold MCP server I can connect?
webmcp.json is a discovery manifest listing public GET resources; there is no tool execution involved. Agents that understand the format use it to find the canonical pages, then read llms.txt and llms-full.txt.
Which file should I use for a quick question?
llms.txt. It is a few kilobytes and answers what WebViewGold is and where the docs live. Use llms-full.txt when the agent has to write code.
How often does llms-full.txt change?
With every documentation update. If you committed a copy, refresh it when you update the WebViewGold project; the grep check above will flag schemes that moved or were renamed.
Can the agent also configure Config.swift and Config.java from the docs?
Yes. The file documents every option with its name and allowed values, so an agent with the WebViewGold project open can set the URL, host, push options and feature flags. Signing and uploads remain manual or go through the Cloud Builder.
Three ways to get your app into the stores
Same result, different amount of work on your side. All three come with the approval guarantee and 24/7 human support.
Do it yourself
$109 per platform, one-time
The WebViewGold license: a finished Xcode or Android Studio project. Paste your URL, set name and icon, switch on features, build and submit. Free updates forever.
See plans on the main pageCloud Builder
$349 per license, one-time add-on
No Mac, no Xcode, no Android Studio: configure in the browser, we build both apps in the cloud and hand them to App Store Connect and the Play Console.
How the Cloud Builder works Try the free demoFull Service
$698 per platform $798
Send us your URL and icon. Our team configures, builds and delivers your app within 72 hours, white-label, with source projects. Both platforms for $1,158.
Full Service details Powered by
Winter Sale prices valid until December 31, 2026.