● Communauté
PleasePrompto
⚡ Instantané
google-ai-mode-skill
par PleasePrompto · PleasePrompto/google-ai-mode-skill
Free Google AI Mode search with citations, driven by Claude — persistent browser profile, query optimization, token-efficient.
google-ai-mode-skill lets Claude query Google's AI Mode (the AI-overview-style answer surface) via a local headless browser with a persistent profile. It returns the answer + citations for further use, optimized prompts on the way in, and minimizes tokens by returning structured JSON. Good alternative when paid search APIs are overkill.
Pourquoi l'utiliser
Fonctionnalités clés
- Zero-config once the browser profile is seeded
- Persistent profile avoids repeated sign-in / consent screens
- Query optimization rewrites your question for AI Mode
- Returns structured answer + citations
- Pure skill (Python browser driver) — no paid API key
Démo en direct
Aperçu en pratique
google-ai-mode-skill.replay
▶ prêt
Cas d'usage
Usages concrets : google-ai-mode-skill
Quand l'utiliser : You want a short, cited answer to a factual question from Google's AI Mode.
Déroulement
-
Seed the profile once
Run the first-time setup so the browser profile is logged in and past the consent screens.✓ Copié
→ Profile dir populated; no further prompts needed
-
Query
Search Google AI Mode: 'What changed in React 19.2 regarding compiler defaults?' — return answer + citations.✓ Copié
→ Short answer + citation URLs
-
Drill deeper
Fetch the top two citation pages and reconcile any contradictions.✓ Copié
→ Reconciled summary
Résultat : A cited answer comparable to a paid search API, at $0.
Pièges
- Profile gets logged out — Re-run the seeding flow; don't share the profile across machines
- Google layout changes and selectors break — Pin the skill version; update when upstream ships a fix
Quand l'utiliser : Before publishing a post, double-check factual claims.
Déroulement
-
Extract claims
From this draft, extract specific factual claims that are worth checking.✓ Copié
→ Claim list with verifiability notes
-
Verify each
Run google-ai-mode on each claim; flag anything unconfirmed.✓ Copié
→ Per-claim status + sources
-
Revise
Edit the draft to correct or caveat unconfirmed claims.✓ Copié
→ Revised draft
Résultat : Content with fewer publishable errors.
Pièges
- Confirmation bias — only checking claims you suspect — Extract all claims, then verify all of them
Combinaisons
Associez-le à d'autres MCPs pour un effet X10
Cache research results as markdown files for reuse
For every question answered today, save a research_<slug>.md in ./notes/.✓ Copié
Inline fact-checking during blog drafting
After each blog section, run AI-Mode fact checks on the claims.✓ Copié
Coût et limites
Coût d'exécution
- Quota d'API
- none (uses Google web surface — respect their ToS and rate limits)
- Tokens par appel
- ~500–2k tokens per answer
- Monétaire
- free
- Astuce
- Cache answers by question hash; don't re-query within a session
Sécurité
Permissions, secrets, portée
Stockage des identifiants : Google session cookies in the persistent profile directory — treat the profile dir as sensitive
Sortie de données : google.com
Ne jamais accorder : shared profile across machines
- Automating search surfaces may be against target-site ToS depending on volume — keep it personal-use and low-rate
- Do not use for anti-bot / CAPTCHA bypass scenarios
Dépannage
Erreurs courantes et correctifs
Selectors no longer match
Update to latest skill version; Google UI changes periodically.
Profile keeps prompting for consent
Re-seed the profile with consent accepted; make sure you keep the profile dir between runs.
Vérifier : ls <profile-dir>
Alternatives
google-ai-mode-skill vs autres
| Alternative | Quand l'utiliser | Compromis |
|---|
| brave-search | You want a first-party search API | Costs money; no ToS ambiguity |
| kindly-web-search | You want a polite, consent-respecting crawler | Different shape |