Skip to content

mcp-integrations-guide.md

Latest commit

 

History

History
232 lines (156 loc) · 8.07 KB

mcp-integrations-guide.md

File metadata and controls

232 lines (156 loc) · 8.07 KB

MCP Integrations Guide

Guía de Model Context Protocol (MCP) integrations que enriquecen skills del venture-studio-toolkit. Las MCP integrations son opcionales — cada skill funciona standalone, pero MCPs específicos desbloquean features adicionales.

⚠️ Estado de implementación (v1.2.0)

Documentación de integration points, NO implementación directa en skills todavía. Cada MCP integration requiere:

  1. User connects MCP en su Claude Code setup (usualmente vía .mcp.json del plugin o claude.ai connectors)
  2. Skill detects MCP availability y usa si disponible
  3. Skill falls back a manual flow si MCP no disponible

v1.3+ rolls out implementation per MCP según demand del dog-food.

MCPs relevantes

Linear MCP (highest impact)

Provider: Oficial Linear MCP via https://linear.app account

Setup: user connects Linear MCP en claude.ai/settings/connectors O en local .mcp.json del plugin

Skills que se benefician:

Skill Enhancement con Linear MCP
accelerator-launchpad Generar SPIKE issues automáticamente para accelerator applications con assignee correcto. Status: recommendation, no implementation yet.
three-horizons Import ventures desde Linear teams/projects en lugar de asking user. Quarterly review comments post-processing.
innovation-scorecard Sync metrics con Linear issue status (feature completion, cycle time).
cost-of-delay-cd3 Import feature backlog desde Linear issues. Rank CD3 results. Update priorities.
improvement-kata Create obstacle issues + daily PDCA sub-tasks en Linear.
vertical-charter Map studio verticals a Linear teams. Sync charter a team description.
sweat-equity-agreement Track vesting milestones en Linear issues.

Funciones MCP típicas usadas:

  • linear.save_issue — crear/update issues con correct assignee
  • linear.list_projects — discover active projects per team
  • linear.list_teams — map studio verticals
  • linear.save_comment — post progress updates

Nota sobre seguridad: user control sobre auth — skills NUNCA access Linear sin explicit user consent.

Context7 MCP

Provider: Context7 (https://context7.com)

Setup: connect Context7 MCP en Claude Code

Skills que se benefician:

Skill Enhancement con Context7
structure-decision Current docs de Stripe Atlas, Firstbase, Clerky para incorporation workflows. Avoid stale snapshots en el skill.
services-hub-setup Latest MSA templates de Cooley Go, Carta. Avoid hardcoded snippets que se vuelven stale.
cap-table-per-venture Current Carta, Pulley API docs si integration futura.
attached-fund-structure Latest Cornerstone LPA v3.X de govclab. Latest SEC Reg D guidance.
accelerator-launchpad Current YC, Techstars, 500 Global application requirements (cambian per cohort).

Funciones MCP típicas:

  • context7.query-docs — get latest docs de service X
  • context7.resolve-library-id — find correct library identifier

Razón: mucha de la info en los skills tiene expiration date (accelerator cohort deadlines, Stripe Atlas pricing, legal template versions). Context7 keeps it fresh.

GitHub MCP (plugin-chrome-devtools-mcp de Anthropic)

Skills que se benefician:

Skill Enhancement
venture-spin-out-playbook Fork/create new repos para new LLC automáticamente post-spin-out.
shared-services-ledger Track intercompany commits (IP transfer docs en git repos).

Funciones:

  • gh CLI vía Bash tool (siempre disponible, no requiere MCP separado)

Slack MCP (optional)

Provider: claude.ai Slack connector

Skills que se benefician:

Skill Enhancement ⚠️ Importante
shared-services-ledger Notify venture owners monthly invoice posted User puede opt-out explicit
innovation-scorecard Notify scorecard updated quarterly Same
improvement-kata Daily check-in reminders a coach + learner Same

Nota importante: algunos users (per feedback memory) explícitamente NO quieren Slack notifications del plugin. Skills que usan Slack MCP deben:

  1. Verificar si user tiene Slack opt-in explícito (vía config notifications.slack: true)
  2. Default OFF
  3. Nunca send sin consent

Figma MCP

Skills que se benefician:

Skill Enhancement
venture-spin-out-playbook Transferir design assets (logo, brand) a new entity's Figma workspace
Cross-plugin con BMT Fase 12 (Branding) Create design systems per venture

Playwright / Chrome DevTools MCP

Skills que se benefician:

Skill Enhancement
accelerator-launchpad Screenshot + analyze accelerator websites para validar application requirements current
jurisdiction-matrix (reference) Validate annual costs by scraping government portals (Delaware Div of Corps, CR Registro Mercantil, etc.) — annual update automation

MCPs NO recomendados / fuera de scope

  • Banking MCPs (Mercury, Stripe): sensitive financial data. Skills no should auto-access accounts. User manages banking manually.
  • DocuSign / Adobe Sign: legal signatures. User responsibility, no automation.
  • Email MCPs (Gmail): LP communications. Too sensitive for automation, user drafts manually.

Patrón de integration recomendado

Cuando un skill usa MCP, seguir este patrón en 4 pasos:

1. Check MCP availability (try detect via tool search o config)

  • Available → use MCP capabilities
  • Not available → fall back a manual flow (pedir user input directly)

2. ALWAYS provide manual fallback (no hard requirement on MCP)

3. Document expected MCP en skill frontmatter:

mcp_recommended:
  - linear       # Enhances venture/team discovery
  - context7     # Current legal template versions
mcp_required: [] # Ningún MCP es hard requirement

4. User can proceed sin MCPs — skill works solo, outputs manualmente.

Roadmap implementation per MCP

MCP Priority Value Target version
Linear Alta Critical for venture discovery + tracking v1.3
Context7 Media Keep legal/accelerator data fresh v1.4
GitHub (vía gh) Ya disponible Native support via Bash v1.0 (existing)
Slack Baja (opt-in) Notifications v1.5
Figma Baja Design asset management v1.6+
Playwright Baja Annual jurisdiction data refresh v1.6+

Setup del usuario (cuando esté disponible)

Cuando un skill con MCP integration se invoque, el plugin:

  1. Lee .mcp.json del plugin root (si existe) para MCPs pre-configured
  2. Detecta user's claude.ai connectors (si running con claude.ai session)
  3. Si MCP target está disponible → usa
  4. Si no → pregunta user si quiere setup OR procede manualmente

User puede gestionar MCPs en:

Contribuciones

Para implementar MCP integration en un skill:

  1. Document en skill frontmatter: mcp_recommended: [linear]
  2. Add try/catch para MCP detection al inicio del skill
  3. Implement MCP-enhanced path (happy case)
  4. Implement manual fallback path (graceful degradation)
  5. Test ambos paths
  6. Update este guide con integration concreta

FAQ

¿Necesito configurar MCPs para usar el plugin?

No. Todos los skills funcionan standalone sin MCPs. Las integrations son enhancements, no requirements.

¿Qué MCPs debería configurar primero?

Para users LATAM serial entrepreneur:

  1. Linear MCP (highest impact si ya usas Linear para tracking)
  2. Context7 (para docs legales current)

¿El plugin guarda MCP credentials?

No. Credentials viven en .mcp.json del user (local) o en claude.ai connectors. Plugin solo invoca MCP funcionalidad — nunca stores auth.

¿Puedo escribir mi propio MCP para integrar con mi sistema custom?

Sí. MCP spec es open — cualquiera puede crear servidor MCP propio. Ver https://modelcontextprotocol.io para documentation. Si tu MCP integra con workflow del venture-studio-toolkit, considerar contribuir integration patterns a este guide.