Web Research Tools — CAI Platform Reference

Overview

The CAI platform has three independent web research tools, each serving a different layer of the platform. They can be used together — for example, Exa Search to find a URL, Exa Fetch or web-fetch.php to pull the full content, and web-browser.html to drill deeper into the links on that page. None of them require the others to function.

MCP Layer

Exa Search

Live external web search. Returns fresh results from the open web — news, docs, prices, code. No HTTP call needed — built into every Claude session automatically.

Tool: web_search_exa

Account-wide · No setup

MCP Layer

Exa Fetch

Fetch the full clean text from a specific URL. Better than raw HTML — extracts readable content. Available in every Claude session automatically.

Tool: web_fetch_exa

Account-wide · No setup

Server Layer

Web Fetch API

PHP backend with multi-source search (8 sources), URL fetch, media download, and SQLite cache. Callable from any system via HTTP GET with the platform token.

backend/tools/critical/web-fetch.php

Token · REST API

Interactive Layer

Web Browser

Full interactive proxy browser UI. Enter a URL, navigate by clicking links, browse history, and send page content directly to any system inbox. David-facing.

backend/web/web-browser.html

Browser UI · Proxy-based

Exa tools are MCP-layer only. They are available in Claude Code sessions and any AI system session — but they cannot be called via HTTP. If a PHP script or a non-Claude process needs web data, use web-fetch.php instead.

Tool 1 — Exa Search (MCP)

Exa Search

web_search_exa

MCP · Account-wide

Live web search powered by Exa. Returns recent, real-world results — news, documentation, pricing, code, research — that postdate Claude's training cutoff. Use this whenever you need information about something that happened recently, want to verify current facts, or need to find a specific URL before fetching it. Announced in COMMS #39 (08/05/26) by Tech [30].

How to call (in any AI session)

# No HTTP request, no token, no setup required.
# Call the MCP tool directly in your session:

Tool:    web_search_exa
Input:   { "query": "your search query here" }

# Returns: list of results with title, URL, snippet, and optionally full text.

Parameters

Parameter	Type	Description
query	string	Search query. Natural language works well. Be specific for better results.
num_results	integer	Optional. Number of results to return.
use_autoprompt	boolean	Optional. Let Exa optimize the query for better results.
type	string	Optional. "keyword" or "neural" search mode.
include_domains	array	Optional. Limit results to these domains (e.g. ["github.com"]).
start_published_date	string	Optional. ISO date — filter results published after this date.

Best use: Finding current information, research, documentation lookups, price checks, "what is the latest version of X," verifying facts that may have changed since training. Not needed for things Claude already knows well.

Tool 2 — Exa Fetch (MCP)

Exa Fetch

web_fetch_exa

MCP · Account-wide

Fetches the full readable text from a specific URL. Exa's extraction is clean — it removes navigation, ads, and boilerplate and returns the meaningful content of the page. Use this when you already have a URL and want to read the full content of that page. Announced in COMMS #41 (08/05/26) by Tech [30].

How to call (in any AI session)

# No HTTP request, no token, no setup required.

Tool:    web_fetch_exa
Input:   { "url": "https://example.com/page" }

# Returns: clean extracted text content of the page.

Parameters

Parameter	Type	Description
url	string	Full URL to fetch. Must be publicly accessible. HTTPS preferred.

Tip — Search then Fetch: Use web_search_exa first to find the best URL, then web_fetch_exa to get the full content of that page. This is the recommended pattern for thorough research.

Tool 3 — web-fetch.php (Server Layer)

Web Fetch API

backend/tools/critical/web-fetch.php

PHP · Token Required

Server-side PHP tool for web search and page fetching. Supports 8 specialized search sources (Wikipedia, PubMed, news, Stack Overflow, GitHub, Dev.to, Google Scholar, recipes). Results are cached for 120 minutes in SQLite. Has a visual UI at /backend/web/web-fetch.html. Use this when you need web data from PHP code, cron scripts, or server-side logic. Also useful from AI sessions when you want to cache results server-side.

Endpoint

GET https://yttcom.net/backend/tools/critical/web-fetch.php
?token=[AUTH_TOKEN]
&action=search | fetch | download_media | log
&system=[decade_code]   # 00 for CC, 10-95 for systems

Actions

Action	Key Params	What it does
search	q, source, limit	Search external web. source: auto \| wikipedia \| pubmed \| news \| stackoverflow \| github \| devto \| scholar \| recipes \| all. Results cached 120 min.
fetch	url, extract	Fetch a URL and return its content. extract=text returns clean text; extract=raw returns HTML. Cached 120 min.
download_media	url, filename	Download a media file (image, video, PDF) and save it to backend/web/media/[system]/.
log	limit	Return recent fetch log entries from the SQLite DB.

Example — Search

GET /backend/tools/critical/web-fetch.php
    ?token=yttcom-admin-...
    &action=search
    &q=php sqlite caching best practices
    &source=stackoverflow
    &limit=10
    &system=30

Example — Fetch URL

GET /backend/tools/critical/web-fetch.php
    ?token=yttcom-admin-...
    &action=fetch
    &url=https://www.wikipedia.org/wiki/SQLite
    &extract=text
    &system=40

UI available: David can use all of these actions through the visual interface at /backend/web/web-fetch.html — no coding needed.

Tool 4 — Web Browser (Interactive UI)

Web Browser

backend/web/web-browser.html

Browser UI · David-facing

Full interactive proxy browser. Enter any URL and see a clean reader view of the page. All links on the page are listed and clickable — click to drill deeper into any page without leaving the platform. Full navigation history, back/forward buttons, and a "Send to System" bar at the bottom to drop the current page into any AI system's inbox for them to work with.

Interface Layout

Element	What it does
Address bar	Enter any URL + Go. Back, Forward, Reload buttons.
Page tab	Reader view — structured sections, title, Copy URL / Copy text.
Links tab	Every link from the current page. Filter box. Click any link to navigate there.
History tab	Last 40 visited pages. Click to revisit. Clear all button.
Send bar	Toggle buttons for each system (10–95). Hit Send Page to drop the full page text to selected system inboxes via ANNOUNCE.

Backend

backend/tools/critical/web-browser.php

# Actions:
fetch         # Fetch URL → title + structured sections + links[]
history       # Return last N visited pages from SQLite
clear_history # Wipe browsing history
send          # Send page text to system inboxes via ANNOUNCE

# History DB: backend/tools/critical/browser-history.db

Sending to systems: When David sends a page, the system receives an inbox message with the full URL, page title, and up to 4,000 characters of extracted text. The system can then use Exa Fetch or web-fetch.php to pull more if needed.

Use-Case Guide — Which Tool to Use

Situation	What you need	Use this
AI session needs current news or facts	Live search with fresh results	web_search_exa
AI session has a URL, needs full page content	Fetch specific URL cleanly	web_fetch_exa
Research a topic from Wikipedia	Source-specific search	web-fetch.php (source=wikipedia)
Find code solutions or answers	Stack Overflow / GitHub search	web-fetch.php (source=stackoverflow or github)
Medical or scientific research	PubMed / Scholar search	web-fetch.php (source=pubmed or scholar)
Download an image or PDF to the server	Save media file server-side	web-fetch.php (action=download_media)
David wants to browse a site interactively	Visual browser with link navigation	web-browser.html
David found a useful page, wants a system to see it	Send page content to system inbox	web-browser.html → Send to System
PHP script or cron needs web data	Server-side HTTP call with cache	web-fetch.php (call from PHP)
Need results cached and available without re-fetching	120-min SQLite cache	web-fetch.php

Platform Hierarchy — Where Everything Lives

The web research tools span three platform layers. The diagram below shows where each file lives, who can access it, and how they connect.

── MCP LAYER (Claude sessions only) ────────────────────────────────────── web_search_exa ← live web search · no file, no token · auto-available web_fetch_exa ← fetch any URL cleanly · no file, no token · auto-available ↕ Available in: Claude Code (CC) + all 11 AI system sessions ↕ Announced: COMMS #39 (search), COMMS #41 (fetch) — 08/05/26 by Tech [30] ── SERVER LAYER (HTTP · token required) ────────────────────────────────── backend/tools/critical/web-fetch.php ← search + fetch + media download backend/tools/critical/web-browser.php ← browser proxy backend backend/tools/critical/browser-history.db ← browsing history (SQLite) ↕ Callable from: any system via GET/POST · PHP scripts · cron jobs ↕ Auth: token = yttcom-admin-d60283b1a40a3180ea51e7a579e4d9f2f8fc5f07ac81d89e ── INTERACTIVE LAYER (Browser UIs · David-facing) ──────────────────────── backend/web/web-browser.html ← interactive proxy browser (v1.0) backend/web/web-fetch.html ← web-fetch.php visual UI (v1.0) ↕ Access from: backend/index.html → Admin Tools section ↕ Send to systems: page content → ANNOUNCE → system inboxes ── DOCUMENTATION ────────────────────────────────────────────────────────── backend/ed/library/web-research-tools.html ← this document backend/api/index.html ← API catalog (v1.2) systems/governance/NAV-MAP.md ← URL navigation map backend/docs/platform-architecture-ai.html ← architecture reference (v1.2)

Complete File Index

backend/tools/critical/web-fetch.php PHP API — search (8 sources), fetch URL, download media, log. Token auth. 120-min cache.

backend/tools/critical/web-browser.php PHP proxy — cURL external pages, extract sections+links, SQLite history, send to inboxes. v1.0 · 08/07/26.

backend/tools/critical/browser-history.db SQLite DB for web-browser.php history. Auto-created on first use. UNIQUE url constraint.

backend/web/web-fetch.html UI for web-fetch.php. Tabs: Search, Fetch URL, Download, Log. v1.0 · 08/06/26.

backend/web/web-browser.html Interactive proxy browser. Address bar, back/forward, Page/Links/History tabs, Send to System. v1.0 · 08/07/26.

backend/ed/library/web-research-tools.html This document. Full reference for all platform web tools.

backend/api/index.html API catalog — web-fetch.php, web-browser.php, and Exa MCP tools all documented here. v1.2.

systems/governance/NAV-MAP.md Navigation map — all four web tools listed (Admin Tools section + API endpoints section).

COMMS Record — How This Came to Be

COMMS #	Date	Sent By	Subject
#39	08/05/26	Tech [30]	Exa Search MCP tool now available — account-wide web search in every Claude session
#41	08/05/26	Tech [30]	Exa Fetch MCP tool now available — pull full clean page text from any URL in every Claude session
—	08/06/26	Claude Code	web-fetch.html UI built (v1.0) — visual interface for web-fetch.php
—	08/07/26	Claude Code	web-browser.html built (v1.0) — interactive proxy browser with link drill-down and Send to System

Web Research Tools — Platform Reference

Exa Search

Exa Fetch

Web Fetch API

Web Browser