Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(x-posts): index the folder and add the MCP registry fleet copy

nirholasf888f4367fcf
diff --git a/docs/x-posts/README.md b/docs/x-posts/README.md
index 71b5703bc..3def4d917 100644
--- a/docs/x-posts/README.md
+++ b/docs/x-posts/README.md
@@ -29,6 +29,7 @@ explicit approval). Each file states what it's for and what has already been ver
| [`threews-avatar-launch.md`](threews-avatar-launch.md) | The three.ws avatar pipeline launch copy |
| [`oracle-trading-mcp-x-post.md`](oracle-trading-mcp-x-post.md) | The oracle trading MCP server launch |
| [`sperax-staking-in-chat-x-post.md`](sperax-staking-in-chat-x-post.md) | SperaxOS chat-native staking, balance, and buy-SPA tool calls. The embedded `/staking` page is not working yet: do not announce until the blockers in the file clear |
+| [`ar-quick-look-fix-x-post.md`](ar-quick-look-fix-x-post.md) | A bug fix, not a feature: `three.ws/ar` and `three.ws/ar/studio` now open real AR on iPhone the way `/avatars/:id/ar` already did |
| [`news-archive-x-post.md`](news-archive-x-post.md) | The news archive |
| [`pumpfun-article-x-post.md`](pumpfun-article-x-post.md) | The pump.fun article |
diff --git a/docs/x-posts/ar-quick-look-fix-x-post.md b/docs/x-posts/ar-quick-look-fix-x-post.md
new file mode 100644
index 000000000..d8a7c0c78
--- /dev/null
+++ b/docs/x-posts/ar-quick-look-fix-x-post.md
@@ -0,0 +1,200 @@
+# X post: AR now actually opens AR on iPhone
+
+Announcement copy for a bug fix, not a new feature: `three.ws/ar` (AR Forge) and `three.ws/ar/studio`
+(AR Studio) now place a model in your real room on iPhone the way `three.ws/avatars/:id/ar` already
+did. Related file: [threews-avatar-launch.md](./threews-avatar-launch.md) sells the avatar-has-a-body
+story; this one sells "AR here means AR, on every phone, not just some of them."
+
+**Thesis of this announcement:** every AR surface on the site claimed to support Quick Look on
+iPhone. Only one of them actually did. The other two silently degraded, in two different ways, and
+nobody who wasn't holding an iPhone would have noticed. That gap is now closed. No roadmap talk, no
+"coming soon."
+
+**Every claim below is true today, verified against the shipped code (not yet deployed, see the
+gate below):**
+
+- Apple Quick Look needs a real USDZ file. `<model-viewer>`'s `ar`/`ar-modes` attribute does **not**
+ generate one from a GLB on its own; it needs `ios-src` pointed at an actual USDZ, or Quick Look
+ never activates.
+- `/avatars/:id/ar` ([src/ar-page.js](../../src/ar-page.js)) always did this right: it fetches the
+ GLB, converts it client-side with three.js's `USDZExporter`, and sets the result as `ios-src`
+ before offering Quick Look.
+- `/ar` (AR Forge) and the `/api/ar` endpoint it falls back to did not. Their `<model-viewer>` had no
+ `ios-src`, so `canActivateAR` was false on every iPhone and the button silently fell back to the
+ plain 3D viewer. A comment in the old code claimed model-viewer converted the GLB automatically.
+ It does not, and never did.
+- `/ar/studio` (multi-model AR) had no Quick Look path at all on iOS. It faked AR with a
+ `getUserMedia` camera feed plus a gyroscope-driven camera look: no plane detection, no correct
+ real-world scale, the model floats instead of resting on your floor. Android got real WebXR
+ tracking through a separate, easy-to-miss "Immersive" button; there was no Scene Viewer fallback
+ for Android devices without WebXR either.
+- The fix reuses the exact pipeline that already worked: a new page,
+ [pages/ar-view.html](../../pages/ar-view.html) +
+ [src/ar-view.js](../../src/ar-view.js), runs the same `usdz-pipeline.js` conversion the avatar page
+ uses. `/api/ar` now hands iOS and desktop off to it (`api/_lib/ar-launch.js`,
+ [tests/ar-export.test.js](../../tests/ar-export.test.js),
+ [tests/api/ar-endpoint.test.js](../../tests/api/ar-endpoint.test.js): 49 passing), and AR Forge
+ inherits the fix with no changes of its own, since its button already fell back to `/api/ar`.
+ Android's Scene Viewer redirect was already correct and is unchanged.
+- AR Studio's control now resolves what the device actually supports and routes there: the immersive
+ WebXR session where it exists, Quick Look on iOS, Scene Viewer on Android otherwise, an honest "open
+ this on your phone" QR prompt everywhere else, each selectable per-model from the selection bar too.
+ Shipped in `2e7a7e9fc`.
+- Shared `/api/ar` links still unfurl with a real render of the model on X, Discord, and iMessage: the
+ redirect to `/ar/view` carries the same `og:image`/title an outright 302 would have dropped.
+
+**Gate before any of this posts: not deployed yet.** All five commits (`2e7a7e9fc` through the docs
+cleanup) are on `main` locally, five ahead of `origin/main`. Per the deploy gate in `CLAUDE.md`,
+shipping to production needs the owner's go-ahead. Do not schedule or post any draft below until
+`https://three.ws/api/version` shows a build that includes this work and `/ar`, `/ar/studio`, and
+`/api/ar?src=...` have been hand-verified on a real iPhone.
+
+**Things to NOT claim:**
+
+1. **Do not say "new AR feature."** Nothing new was added to what AR Forge or AR Studio could already
+ do on Android or in the browser. The whole story is that iPhone now gets what was always promised.
+2. **Do not name the old bug's exact symptom as if it were widely reported.** Frame it as "we found
+ this ourselves and fixed it," not "users complained." Nobody publicly complained; it was caught in
+ review.
+3. **Do not claim AR Studio's iPhone experience places multiple objects with real tracking.** Quick
+ Look places one object at a time by design (a USDZ is a single scene). Multi-object placement with
+ real tracking is still Android/WebXR only; the per-model "Place in your space" button on iOS is
+ correct and true, but do not stretch it into "full multi-model AR on iPhone."
+
+---
+
+## 1. Main post (recommended)
+
+258 characters with X's 23-character URL count.
+
+> We shipped three AR surfaces. Only one of them actually opened AR on an iPhone.
+>
+> The other two quietly fell back to a flat 3D view or a fake camera overlay. Nobody would've
+> noticed unless they were holding an iPhone.
+>
+> Fixed. All three now use real Quick Look.
+>
+> three.ws/ar
+
+Why this one: the honesty is the hook. Admitting a real gap and naming the fix precisely reads as
+credible in a way "AR now works!" never does, and it invites the reply "wait, which two."
+
+## 2. Alternate: the technical explainer
+
+269 characters.
+
+> Apple Quick Look needs a real USDZ file. `<model-viewer>` does not generate one from a GLB on its
+> own, no matter what the docs imply.
+>
+> Two of our AR pages never set it. iPhone silently got a downgrade nobody could see from the code.
+>
+> Fixed, verified on-device.
+>
+> three.ws/ar
+
+Why this one: for a technical audience, naming the exact mechanism (`ios-src`, no auto-conversion) is
+more convincing than any adjective. It also preempts "how do you know it's fixed."
+
+## 3. Alternate: the room-floor angle
+
+264 characters.
+
+> AR that doesn't know where your floor is isn't AR. It's a video filter.
+>
+> Our multi-model AR studio faked it on iPhone with a camera overlay: no real scale, nothing to rest
+> the model on.
+>
+> Now it hands off to Apple's own Quick Look. Real tracking, real floor.
+>
+> three.ws/ar/studio
+
+Why this one: leads with a strong, quotable claim about what AR even means, which travels further
+than a bug-fix framing on its own.
+
+## 4. Alternate: the short one
+
+Under 200 characters. Good as a reply-chain opener or a standalone if the thread below feels heavy.
+
+> Fixed: `three.ws/ar` and `three.ws/ar/studio` now open real Apple Quick Look on iPhone, the same
+> way our avatar pages always have.
+>
+> No more silent fallback to a flat 3D view.
+>
+> three.ws/ar
+
+---
+
+## Thread version
+
+Post 1 is the main post above. Then:
+
+**2/**
+
+> Quick Look needs a USDZ file, not a GLB. `<model-viewer>`'s AR attribute offers Quick Look as a
+> mode, but it will not convert the model for you: you have to hand it a real USDZ as `ios-src`, or
+> the button quietly does nothing useful.
+
+**3/**
+
+> Our avatar AR page (`/avatars/:id/ar`) always did this: fetch the GLB, convert it to USDZ in the
+> browser with three.js's own exporter, hand it to Quick Look. That page was always correct.
+>
+> AR Forge and AR Studio were not built the same way.
+
+**4/**
+
+> AR Forge fell back silently to the plain interactive 3D viewer on iPhone. No error, no warning: the
+> button just stopped being an AR button.
+>
+> AR Studio went further: it faked AR with your camera feed and the phone's gyroscope. The model
+> floats. Nothing anchors it to your floor. It looks like AR in a screenshot and nothing like it in
+> your hand.
+
+**5/**
+
+> Both are fixed the same way: a shared page now runs the exact USDZ conversion the avatar page
+> always used, and both surfaces hand off to it on iPhone.
+>
+> Android's Scene Viewer redirect was already correct the whole time and didn't need touching.
+
+**6/**
+
+> Try it: three.ws/ar, or forge a scene at three.ws/ar/studio and tap the AR button.
+>
+> three.ws
+
+---
+
+## Replies worth pre-writing
+
+**"How did this ship broken in the first place?"**
+
+> `<model-viewer>`'s AR attributes make Quick Look look wired up even when it isn't: the button
+> renders, the attribute is set, and the only visible symptom is that tapping it on an iPhone does
+> nothing useful. It doesn't throw an error. Caught in review, not from a report.
+
+**"Does this mean AR Studio now places multiple objects in real tracked AR on iPhone?"**
+
+> No, and we're not claiming that. Quick Look places one object at a time; that's how USDZ works.
+> What's fixed is that placing one is now real device AR instead of a camera overlay. Multi-object
+> placement with true tracking stays an Android/WebXR thing for now.
+
+**"Was any user-facing data affected?"**
+
+> No. This is a rendering/AR-launch bug, not a data issue. Nothing about accounts, wallets, or stored
+> models was involved.
+
+---
+
+## Notes on framing
+
+- **Lead with the honesty, not the fix.** "We found a real gap and closed it" earns more trust than
+ "new AR upgrade," and it's the more accurate description of what happened.
+- **Do not mention $THREE.** This is a reliability fix, not a growth feature; forcing a coin mention
+ onto it would read as opportunistic.
+- **Screenshots, if used, should be before/after on the same physical spot**, the way
+ [3d-ar-studio-launch.md](./3d-ar-studio-launch.md) already frames its media: a real floor in frame
+ is the whole point, so a screenshot without one undercuts the post.
+- **Do not combine this with the AR Studio open-source package announcement**
+ ([3d-ar-studio-launch.md](./3d-ar-studio-launch.md)). That post is about giving the surface away as
+ a library; this one is about a bug on the site everyone already uses. Mixing them muddies both.
diff --git a/docs/x-posts/mcp-registry-fleet-x-post.md b/docs/x-posts/mcp-registry-fleet-x-post.md
new file mode 100644
index 000000000..225bb6c04
--- /dev/null
+++ b/docs/x-posts/mcp-registry-fleet-x-post.md
@@ -0,0 +1,110 @@
+# X post: 72 MCP servers in the official registry
+
+Announcement copy for the three.ws MCP fleet as a whole, rather than any single server. The news
+is the shape of the thing: one namespace, `io.github.nirholas`, carrying 72 distinct servers in the
+official Model Context Protocol registry, every one of them at its current published version.
+
+**Thesis of this announcement:** we did not ship one MCP server and call it a platform. The agent
+tooling is a fleet, it is all installable in one command each, and it is all current. Nothing else.
+No roadmap, no coin talk.
+
+**Every claim below is true today**, verified against registry.modelcontextprotocol.io on
+2026-08-20: 72 distinct servers under `io.github.nirholas`; 36 of them published or version-bumped
+in a single run today; zero remaining version gaps between the shipped npm packages and their
+registry entries; every stdio server installs with `npx -y @three-ws/<name>`; the newest entry is
+`io.github.nirholas/[email protected]`, live and flagged latest.
+
+**Things to NOT claim:**
+
+1. **Do not claim a rank, and do not say "most of any publisher".** A partial sweep of the registry
+ (3,176 servers) put the largest namespace observed at 213, so 72 is a lot and is not a record.
+ The number is impressive without a superlative, and a superlative is exactly what someone will
+ spend ten minutes disproving.
+2. **Do not imply all 72 are equally mature.** They range from load-bearing servers to small
+ single-purpose ones. "72 servers" is true; "72 production platforms" is not.
+3. **Do not quote a download or usage figure.** We have not measured one.
+
+---
+
+## 1. Main post (recommended)
+
+268 characters with X's 23-character URL count.
+
+> 72 MCP servers, one namespace, all live in the official registry.
+>
+> Wallets, payments, 3D avatars, on-chain identity, market data, x402. Each installs into Claude or
+> Cursor in one command.
+>
+> We have been building the agent toolchain, not a demo.
+>
+> three.ws
+
+Why this one: the number does the work in the first six words, the second line proves it is a range
+rather than 72 wrappers on the same API, and the last line is the only opinion in the post. It reads
+as a status report, which is the tone that survives scrutiny.
+
+## 2. Alternate: the builder cut
+
+> Every three.ws MCP server is now current in the official MCP registry. 72 of them.
+>
+> ```
+> npx -y @three-ws/onchain-agent-wallets
+> npx -y @three-ws/metaplex-agent-mcp
+> npx -y @three-ws/x402-mcp
+> ```
+>
+> Agent wallets, on-chain deploy, and x402 payments. Pick one and it runs.
+
+Three real commands beat any adjective. All three are verified working.
+
+## 3. Alternate: the quiet flex
+
+> Shipped 36 MCP server updates to the official registry today.
+>
+> Not a launch. Just the fleet catching up to the code.
+>
+> 72 servers total, all current.
+>
+> three.ws
+
+For an audience that reads restraint as competence. Works best as a reply to your own earlier
+launch post rather than standalone.
+
+---
+
+## Thread version
+
+Post 1 is the main post. Then:
+
+**2/**
+
+> The registry entry is the boring part that matters. It means an agent client can discover the
+> server, see what it needs, and install it without a human copying a JSON blob out of a README.
+
+**3/**
+
+> What is in there: agent wallets with on-chain spending limits, x402 payment rails, Solana and EVM
+> tooling, on-chain agent identity, 3D avatar generation, market data, notifications, and the
+> account plumbing underneath all of it.
+
+**4/**
+
+> Every one is self-custodial where money is involved. No server holds a key. The ones that move
+> funds show you the transaction and refuse to broadcast without an explicit yes.
+
+**5/**
+
+> The newest is onchain-agent-wallets: give an agent a Solana spending allowance instead of a
+> private key, and the token program enforces the ceiling.
+>
+> github.com/nirholas/onchain-agent-wallets
+
+---
+
+## Notes on framing
+
+- **Lead with the number, not with "MCP".** The count is legible to anyone; the acronym is not.
+- **Name concrete categories.** "72 servers" invites the reply "72 of what". Answer it in the post.
+- **No $THREE in this one.** The fleet story stands on its own, and the coin belongs in posts where
+ a fee or a holder benefit is actually wired, like the metaplex deployer copy.
+- Registry namespace, for anyone who asks: `io.github.nirholas`.