three.ws Herald
io.github.nirholas/herald-mcp · 231 stars · JavaScript · Apache-2.0
MCP server Tell your human in person: a 3D character walks onto their open browser tab and says it out loud.
Install
The repo has no one-line install. Follow its README.
Published as
- npm
@three-ws/herald-mcpstdio
From the server's entry in the official MCP Registry.
Configuration
| Name | Set as | What it is |
|---|---|---|
THREE_WS_API_KEYRequired · Secret | Env var | A three.ws API key (sk_live_… / sk_test_…) with the herald:announce scope, or an OAuth access token for the same account, carried as Authorization: Bearer. Required: an announcement is always delivered to the key owner's own sessions. Aliases: THREE_WS_TOKEN, THREE_WS_BEARER. Treat like a password. |
THREE_WS_BASEOptional | Env var | Base URL of the three.ws API. Override only when self-hosting or targeting a preview deployment. |
THREE_WS_TIMEOUT_MSOptional | Env var | Per-request timeout in milliseconds for the announce call. |
Files
@three-ws/herald-mcp
Let your agent tell you something in person.
An MCP server with one idea behind it: when an agent finishes something that mattered, or gets stuck, it should not add a line to a log you will never scroll back to. It should walk up to you and say so.
Wire this in, and it does. Your own 3D companion appears in the corner of whatever browser tab you have open, gestures, and says the line out loud, with a link to click through.
you: run the migration, then the backfill, and tell me when it's done
agent: [runs for 22 minutes]
[calls announce_result: task "the backfill", outcome succeeded, 1320s]
your browser: your avatar walks on: "the backfill finished in 22m"
Tools
| Tool | What it does |
|---|---|
announce |
Say one line, with importance (0-100), tone, a gesture, and an optional link. |
announce_result |
Report a finished task. The urgency is chosen from the outcome, so a failure cuts through quiet hours and a success does not. |
check_rail |
Prove the key and the rail work, without interrupting anybody (queues at importance 0, which every client drops). |
Install
npx -y @three-ws/herald-mcp
Claude Code:
claude mcp add herald -e THREE_WS_API_KEY=sk_live_... -- npx -y @three-ws/herald-mcp
Claude Desktop / Cursor (mcp.json):
{
"mcpServers": {
"herald": {
"command": "npx",
"args": ["-y", "@three-ws/herald-mcp"],
"env": { "THREE_WS_API_KEY": "sk_live_..." }
}
}
}
The key
Create one at three.ws/dashboard/developers
with the herald:announce scope, and put it in THREE_WS_API_KEY.
An announcement is always delivered to the key owner's own live sessions. There is no recipient parameter: this server cannot be pointed at anybody else, so the worst a leaked key can do is annoy the person who leaked it. Revoke it in the same place.
| Variable | Required | Default | Notes |
|---|---|---|---|
THREE_WS_API_KEY |
yes | API key with herald:announce, or an OAuth access token. Aliases: THREE_WS_TOKEN, THREE_WS_BEARER. |
|
THREE_WS_BASE |
no | https://three.ws |
Override for self-hosting or a preview deployment. |
THREE_WS_TIMEOUT_MS |
no | 20000 |
Per-request timeout. |
When to use it (and when not to)
Announcements are scarce on purpose. The receiving client applies an importance floor, a rate limit, dedupe, quiet hours, and a freshness window, so anything that does not clear the bar is dropped rather than queued. That is the feature: an interruption only works while it is rare.
Good: a long job finished, a job failed, a decision is needed, something broke, money moved.
Bad: progress updates, "starting step 3 of 7", anything the person can read later. Those belong in the notification inbox (@three-ws/notifications-mcp).
If nothing is open to hear it, the line expires in about five minutes. The rail is a live channel, not an archive.
Example calls
// something broke, and it should wake somebody up
{ "name": "announce", "arguments": {
"text": "The payments worker has been failing for 10 minutes",
"from": "watchdog", "importance": 95, "tone": "error",
"url": "https://three.ws/dashboard", "key": "payments-worker-down" } }
// a long task finished, and it should not
{ "name": "announce_result", "arguments": {
"task": "the nightly backfill", "outcome": "succeeded", "seconds": 1320 } }
// wiring check
{ "name": "check_rail", "arguments": { "note": "claude-code" } }
How it reaches a browser
POST /api/herald/announce queues the line on the account's rail; the page
listens over SSE and hands it to @three-ws/herald,
which decides whether it clears the person's own rules and then delivers it
through the 3D companion (or an accessible card when there is no GPU).
Try the whole loop, including the rules engine, at three.ws/herald.
License
Apache-2.0.
Facts
- Kind
- MCP server
- Group
- Uncategorized
- Stars
- 231
- License
- Apache-2.0
- Language
- JavaScript
- Last push
- 2026-10-12
- Forks
- 63
- MCP Registry
- io.github.nirholas/herald-mcp
- Homepage
- three.ws
- Topics
- animation, avatar, gltf, llm, mcp, webgl, 3d-generation, ai-agents, digital-human, generative-ai, glb, image-to-3d, mcp-server, model-context-protocol, solana, text-to-3d, threejs, web-components, webxr, x402
- 1REAio.github.morluto/reaReverse engineer anything from your terminal or agent with one CLI and MCP server.125.3k
- 2com.browser-use/browser-usecom.browser-use/browser-useControl a real Chrome browser to complete any task: fill forms, extract data, book flights.117.7k
- 3World Monitorapp.worldmonitor/mcpLive markets, conflicts, country risk, chokepoints, energy, and China decision signals. 93 tools.88.3k
- 4Scrapling MCP Serverd4vinci/scrapling🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scraplingdev86.9k
- 5Everythingmodelcontextprotocol/serversThis MCP server attempts to exercise all the features of the MCP protocol. It is not intended to be a useful server, but rather a test server for builders of MCP clients. It implements prompts, tools, resources, sampling, and more to showcase MCP capabilities.85.8k
- 6Fetchmodelcontextprotocol/serversA Model Context Protocol server that provides web content fetching capabilities. This server enables LLMs to retrieve and process content from web pages, converting HTML to markdown for easier consumption.85.8k