What woke up overnight.

A new layer: plain-English WebMCP education, a live experiment pipeline, and a way to share what we find without pretending we built it.

The immediate target is imperative-first WebMCP: tools registered through document.modelContext.registerTool, with schemas, lifecycle, annotations, and verification.

Open the SYNAL WebMCP test harness

Explore carefully. Keep the evidence.

THE HUMAN OUTCOME

Explore carefully. Keep the evidence.

The lab distinguishes discovered projects from tested tools, records provenance and makes permissions and limitations visible.

The new layer

WebMCP explainer

Plain-English education on WebMCP: imperative tools, registration, schemas, lifecycle, annotations, and verification.

OpenAI WebMCP plugin

Captures the ChatGPT integration’s imperative-first target and maps it to SYNAL’s capability contract.

Ecosystem sweep lenses

Seven repeatable lenses: discovery, builder, architecture, trust, commercial, future signals, and T4H opportunity.

WebMCP top 10 watchlist

The first ten projects, extensions, userscripts, bridges, and directories worth trying or sharing.

Experiment protocol

Green, amber, and red lanes so rough experiments can run without endangering the main browser, credentials, or data.

Sharing & credit protocol

Rules for crediting creators, recording licences, labelling discovery versus testing, and separating T4H-built from shared work.

First education posts

Six education pieces ready to draft and publish.

The important reframe

A tool registration is not success. An agent invocation is not success.

Success is a real action, a visible state change, a truthful result, and an evidence receipt.

That gives SYNAL its first technical position:

SYNAL helps websites expose safe, structured, task-completing capabilities to AI agents — and proves what actually happened.

Top 10 experiment watchlist

Everything below is labelled DISCOVERED / NOT TESTED. Nothing is claimed as installed, run, or verified.

1. GoogleChromeLabs/webmcp-tools

Inspector, evaluations, polyfill, and demos. The reference toolkit.

2. TueJon/webmcpify

App inventory, tool proposal, integration, and repair skill.

3. WebMCP-org/webmcp-userscripts

Agent-capable existing sites without changing the site owner’s code.

4. Cloudflare Agents SDK WebMCP adapter

The most relevant bridge to T4H’s Cloudflare and MCP direction.

Red lane: Tampermonkey MCP

Explicitly red lane because it can write, patch, and delete userscripts.

Each item has a risk lane, required environment, evidence requirement, and stop conditions.

Experiment lanes

Green lane

Read-only, isolated, reversible experiments. No credentials. No production data. No consequential actions.

Amber lane

Controlled experiments with explicit approval, bounded scope, logging, and a defined rollback path.

Red lane

Write-capable, privileged, or hard-to-reverse experiments. Requires a separate profile, explicit approval, evidence capture, and stop conditions.

Sharing without pretending

Credit the creator. Link the original work. Record the licence.

Label every item DISCOVERED, TESTED, ADAPTED, RECOMMENDED, or NOT RECOMMENDED.

Separate what T4H built, adapted, funded, shared, or merely found. Never convert someone else’s project into implied T4H IP.

This lets SYNAL become a credible discovery and education surface immediately — before T4H has shipped its own WebMCP tools.

First six education posts

  1. WebMCP explained: websites can expose real tools to AI agents.
  2. The difference between an agent clicking and a capability completing.
  3. Ten WebMCP experiments worth watching.
  4. How to make an existing website agent-capable without rebuilding it.
  5. Why annotations are not security boundaries.
  6. The SYNAL capability contract: declare, discover, authorise, execute, verify.

Still open

The legal and operating documentation gap remains open: Terms of Use, Privacy Policy, Builder Agreement, Marketplace Terms, AI Use Policy, Security Policy, and incident-response documentation still need to be collected or drafted.

But the position has changed: SYNAL now has a frozen evidence baseline, a clear WebMCP education narrative, a live experiment pipeline, and a sharing model that can start tomorrow without overclaiming.

Talk to us