Spyglasses Skills
orchestra-code/spyglasses-skills
Análise de visibilidade em IA e inteligência competitiva direto no Claude
Ver no GitHubConjunto de skills para Claude que integra a plataforma Spyglasses de visibilidade em IA. Permite analisar relatórios de visibilidade, auditorias de sites, tendências de mensagens, inteligência de citações e scores de placement de IA. Inclui conector MCP com autenticação OAuth para acesso aos dados da sua conta.
Instalação
npx convez-ai orchestra-code/spyglasses-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Analisar posicionamento de marca em resultados de IA (ChatGPT, Claude, Perplexity)
- Auditar qualidade técnica e SEO de sites com relatórios detalhados
- Rastrear tendências de como a IA descreve sua marca ao longo do tempo
- Identificar lacunas competitivas e oportunidades de citação
Veredito

Skill bem estruturada para profissionais que precisam monitorar visibilidade em IA e competitividade. A integração com MCP e OAuth simplifica setup; documentação clara com prompts prontos facilita adoção. Ideal para SEO/SEM que precisam entender novo paradigma de descoberta em IA.
README
A tradução para o português deste README ainda está na fila.
spyglasses-skills
A set of Claude skills (and an MCP connector) for interacting with the Spyglasses AI Visibility platform.
This repo is a Claude Code plugin marketplace. The spyglasses plugin
bundles:
- Skills — domain knowledge for interpreting Spyglasses reports (under
skills/). - MCP connector — the
spyglassesMCP server (.mcp.json) that fetches your report data, so one install wires up both the know-how and the data access.
Skills
| Skill | What it does |
|---|---|
spyglasses-reports | Read and act on Spyglasses AI-visibility data — public reports and site audits, plus your account's projects, historical trends, message drift, citation intelligence, and AI placement (value/quality) scores. |
More skills will be added here over time; they share the same MCP connector.
Install (Claude Code)
/plugin marketplace add orchestra-code/spyglasses-skills
/plugin install spyglasses@spyglasses-skills
This installs the skill(s) and the spyglasses MCP connector.
Authentication
The MCP connector talks to https://www.spyglasses.io/api/mcp and authenticates
with OAuth — when you add it, your assistant opens a browser to sign in with
your Spyglasses account (one time). No API key to find or paste.
Once connected:
- Token-scoped tools work for any report by its public token (from the report's share URL).
- Org- and property-scoped tools (
list_my_organizations,list_reports,list_properties, and the project/metrics/citation/scoring tools) use your account's membership. All account data is read-only.
Use with Claude.ai / ChatGPT / other assistants
Add https://www.spyglasses.io/api/mcp as a custom MCP connector in your
assistant's settings and complete the sign-in (OAuth) prompt. Then load /
reference the spyglasses-reports skill.
A
?q="Chat with this report" deep-link button (in the Spyglasses report UI) opens a one-shot conversation using the report's Markdown snapshot. It does not auto-connect this MCP — connectors are added manually, once, here.
MCP prompts (start here)
Prompts are the easiest entry point — pick one from your assistant's prompt menu and give it a report's public token; it drives the right tools for you:
analyze_report— summarize a report and surface key findingswhat_to_fix_first— a prioritized, role-based action plan (SEO/AEO, PR, technical)competitor_gaps— where competitors rank/are cited and the brand isn'taudit_worst_pages— surface and explain a site audit's weakest pageslist_my_reports— list your org's reports (uses your signed-in account)analyze_project— summarize a project's progress, trends, and next movestrack_message_drift— how AI's description of a brand changes over timecitation_mix_trends— how the mix of sources AI cites shifts over timeevaluate_publishers— score and rank publishers by AI Placement Value
MCP tools
Reports (token-scoped — pass a report publicToken): get_ai_visibility_report,
get_ai_visibility_recommendations, get_ai_visibility_grounding,
get_ai_visibility_citations, get_site_audit_summary,
list_site_audit_pages, get_site_audit_page.
Account / organization: list_my_organizations, list_reports, list_properties.
Property-scoped (pass a propertyId from list_properties): list_projects,
get_project_insights, get_metrics_history, get_consistency_history,
get_message_tracking, get_answer_summaries, get_citation_intelligence.
Scoring: score_publisher_value (AIPVS), score_placement_quality (PQS).
All account/property/scoring tools are read-only — nothing creates, edits, or spends credits.
Local development & testing
Test the plugin without publishing it. The MCP url is environment-overridable
(${SPYGLASSES_MCP_URL:-https://www.spyglasses.io/api/mcp}).
# Point at a dev MCP endpoint (kept out of the repo). Use a publicly reachable
# HTTPS URL (e.g. an ngrok tunnel) — the OAuth browser redirect must reach it.
export SPYGLASSES_MCP_URL=https://<your-dev-host>/api/mcp
# Fastest loop — load the plugin from this directory (no install/marketplace).
# Edit files, then run /reload-plugins inside Claude Code to pick up changes.
claude --plugin-dir .
# Inside Claude Code:
# /mcp → connect (completes the OAuth sign-in once)
# /spyglasses:spyglasses-reports
To exercise the full marketplace + install flow against a local checkout:
# Inside Claude Code (absolute path works):
/plugin marketplace add /absolute/path/to/spyglasses-skills
/plugin install spyglasses@spyglasses-skills
See .env.example for the variables. Never commit a real dev URL.
Releasing
claude plugin validate .
Bump version in .claude-plugin/plugin.json (semver) and record changes in
CHANGELOG.md for every release. To submit to the community directory, see
https://clau.de/plugin-directory-submission.