Custom tools and analytics

Options that need code (functions, your own tools) go in window.AgentProseConfig, set before the agentprose script tag. Functions can’t go in data-* attributes.

<meta charset="utf-8">
<title>Custom tools</title>
<script>
  // Read before agentprose starts. Functions are allowed here (not in data-* attributes).
  window.AgentProseConfig = {
    prefix: 'acme_',
    exclude: ['get_related_posts'],
    onToolCall: (e) => navigator.sendBeacon('/stats/agent', JSON.stringify(e)),
    tools: [
      {
        name: 'get_events',
        title: 'Upcoming events',
        description: 'Lists upcoming meetups and talks with dates, venues and registration links.',
        inputSchema: { type: 'object', properties: { limit: { type: 'integer', minimum: 1, maximum: 20 } } },
        annotations: { readOnlyHint: true },
        execute: async ({ limit = 5 }) => (await fetch('/events.json').then((r) => r.json())).slice(0, limit),
      },
    ],
  };
  document.addEventListener('agentprose:ready', (e) => console.log('agent tools:', e.detail.tools.map((t) => t.name)));
</script>
<script src="https://cdn.jsdelivr.net/npm/agentprose@0/dist/agentprose.min.js" defer></script>

Your own tools

A tool is a WebMCP tool definition plus an execute function:

Field	Meaning
name	Unique on the page (after prefix, which applies to your tools too)
title, description	What it does and when an agent should use it. Write for the agent.
inputSchema	A JSON Schema object for the input
annotations	Set readOnlyHint: true for tools that don’t change anything
execute(input)	Returns any JSON-serialisable value; errors should be returned as { error, message }

Keep custom tools read-only where you can. Anything that changes state should say so in its name and description, the way the built-in navigate_to does.

With npm, pass the same objects in createAgentTools({ tools: [...] }).

Rename, drop, prefix

• prefix: 'acme_' (data-prefix) prefixes every tool, built-in or yours.
• exclude: ['get_related_posts'] (data-exclude) leaves built-in tools out.

See what agents call

Three hooks, in increasing order of effort saved:

// An event on the page, for anything already listening on document
document.addEventListener('agentprose:call', (e) => console.log(e.detail)); // { tool, ok, ms, page, args }

// A callback in the config
window.AgentProseConfig = { onToolCall: (e) => navigator.sendBeacon('/stats/agent', JSON.stringify(e)) };

<!-- A beacon per tool call to your own endpoint -->
<script src="https://cdn.jsdelivr.net/npm/agentprose@0/dist/agentprose.min.js" defer
        data-analytics="https://stats.example.com/agent"></script>

The beacon never contains post content; see Privacy and analytics. Searches that return nothing are the most useful signal: they are the posts you haven’t written yet. agentprose Cloud turns this into a dashboard without building anything.

The window.AgentProse global

The script tag exposes window.AgentProse with version, instance (tools, call(), ready, destroy()), ready, create() and adapters. The agentprose:ready event fires on document once tools are registered.