v1.4.0 — Agent Mode: MCP endpoint, deep links, and a visual upgrade

Live: https://navigatorslab.com/tools/ · Free & open source (MIT) — no accounts, no uploads, nothing retained.

🤖 Agent Mode (the headline)

The suite now serves AI agents as first-class users:

• MCP endpoint — POST /tools/mcp speaks Model Context Protocol (JSON-RPC 2.0, Streamable-HTTP-style, stateless, no auth). Four compute tools, all pure & unit-tested: nl_catalog, qr_payload (Wi-Fi/vCard/mailto/text with spec-correct \ ; , : escaping), text_diff (LCS), text_stats. GET→405, Origin allow-list, security headers stamped on every response — verified live on production.
• Deep links — every tool accepts strict allow-listed URL parameters: qr.html?text=…&ec=M renders with zero clicks, shrink.html?url=…&format=webp&targetKB=300 pre-loads a same-origin file and pre-sets the target, sign.html?url=…&date=1, ocr.html?url=…&autostart=1, textdiff.html?a=…&b=… auto-compares, and more. ?url= is same-origin only, enforced in code; applied params surface an "Agent mode" chip.
• Machine-readable docs — llms.txt + llms-full.txt at /tools/ and the site root, tools.json, and a full reference page at /tools/agents.

🎨 Visual upgrade

• Aurora hero (layered radial gradients + masked grid) with a gradient-headline
• Tool cards: gradient icon tiles, hover lift + glow, scroll-reveal entrance (reduced-motion aware)
• New "for AI agents, too" hub section with pill, CTA and cards — localized EN/TR/DE
• Table links darkened to pass WCAG 4.5:1 (agents page a11y 91 → 100)

✅ Verification

• 76/76 unit tests (17 new MCP tests: initialize/tools/list/tools/call, error codes, WIFI escaping, diff & stats engines)
• 27/27 E2E checks (4 new: agents page + llms served, ?text= auto-render, ?url= data: analysis, params pre-set)
• 25/25 security checks across all 17 pages · 0px mobile overflow (agents included)
• Lighthouse 97–100 on audited pages · CI green · production MCP round-trip + deep-link render verified in a real browser