Bottom Line
The plugin ecosystem is young and has zero connectors for our specific tools (Kit, Supabase, Railway, Resend). The marketplace plugins are mostly general-purpose productivity tools — useful but not transformative. The real opportunity is building 2-3 custom skills that encode Charlotte's and Stitch's session processes, which would eliminate the ~15 minutes of manual checklist execution at every session open/close. However, this is a "nice to have" — not now. We have a live production bug, undeployed scorecard, and a March 5 Kit deadline. Revisit after launch stabilizes.
1. What I Found: MCP Connector Registry
I searched the MCP connector registry for every tool in our stack. Every single search returned zero results.
| Search Terms | Results |
|---|---|
| kit, convertkit, email marketing | 0 |
| stripe, payments, billing | 0 |
| supabase, database, postgres | 0 |
| railway, deployment, hosting | 0 |
| resend, transactional email | 0 |
| instagram, meta, social media | 0 |
| vercel, frontend deployment | 0 |
| github, git, version control | 0 |
| google drive, docs | 0 |
The registry is either very new, very small, or our stack is too niche. Either way, there are no drop-in connectors that would let Charlotte or Stitch interact with Kit/Supabase/Railway directly from within a Cowork or Claude Code session today.
2. What I Found: Plugin Marketplace
Anthropic's knowledge-work-plugins collection has ~15 plugins across categories. The ecosystem reportedly has 9,000+ plugins total (as of Feb 2026), but most are community-contributed and unvetted. Here's what's potentially relevant to us:
| Plugin | What It Does | Relevance |
|---|---|---|
| productivity | Task management, scheduling, workflow automation | Medium |
| marketing | Content creation, campaign management, analytics | Medium |
| sales | Pipeline, outreach, CRM workflows | Low |
| engineering | Code review, CI/CD, dev workflows | Medium |
| finance | Invoicing, expense tracking, financial analysis | Low |
| operations | Process automation, compliance, documentation | Medium |
| cowork-plugin-management | Create and manage custom plugins from within Cowork | High |
| customer-support | Ticketing, knowledge base, response templates | Low |
| data | Data analysis, visualization, ETL | Low |
| design | Design workflows, asset management | Low |
The "Medium" ratings don't mean these plugins are bad — they're general-purpose tools that might save a few minutes here and there. But none of them know about our session processes, brand rules, DECISIONS.md, or multi-agent coordination. They'd be marginal improvements at best.
3. Custom Skills: Where the Real Value Is
After reading the skill-creator documentation, the highest-impact opportunity is packaging our existing manual processes as reusable skills. Right now, Charlotte's opening process is an 8-step checklist in CLAUDE.md that gets executed manually every session. A custom skill could encode this as a single invocable command with built-in error handling and verification.
High-Value Custom Skill Candidates
Reads values.md, session-context.md, latest handover, DECISIONS.md, inbox, runs conflict checks, platform risk check, decision gate check — then outputs a structured briefing. Would save ~10 minutes per session open and eliminate missed steps (like the Session Log gap we caught this session).
Updates session-context.md, writes handover doc from template, logs decisions, updates session log, runs brand file review, marketing consistency audit, archives inbox items, generates Jason briefing. Would save ~10 minutes per session close and ensure nothing gets skipped.
Takes content + content type as input, generates relevant expert personas, runs scoring, flags if below 7.0, applies top recommendations, outputs final scored version. Would standardize quality and make the process faster to invoke.
Scans content against brand rules (no "AI", correct tiers/pricing, time claims, founder story accuracy). Useful but lower frequency than session processes. Could be a sub-step inside session-close instead of standalone.
One-command launch of Sage/Lux/Bolt/Mint subagents with pre-loaded context from DECISIONS.md and session-context.md. Nice but we don't run advisory analyses often enough to justify the build time right now.
What About Marketplace Plugins?
Could complement our task management (inbox system). Worth trying after launch to see if it adds anything beyond what our inbox folder already handles.
Too generic for our needs. Charlotte already handles marketing strategy with brand files + Expert Panel. We don't have customers yet for support/sales tooling. Finance is premature pre-revenue.
4. Build Effort Estimate
Based on the skill-creator documentation, each custom skill requires a SKILL.md file (the instruction set), optional bundled resources, and testing/evaluation. The skill-creator skill itself can help build and test these.
| Skill | Effort | Time Savings per Use | Frequency |
|---|---|---|---|
| session-open | ~30 min to build + test | ~10 min saved | Every session (daily+) |
| session-close | ~30 min to build + test | ~10 min saved | Every session (daily+) |
| expert-panel | ~20 min to build + test | ~5 min saved | 2-3x per week |
| brand-audit | ~15 min to build | ~3 min saved | Bundled in session-close |
The session-open and session-close skills would pay for themselves within 2-3 sessions. But the build time competes with higher-priority work right now.
5. Charlotte's Recommendation
Not Now — Revisit After Launch Stabilization
Here's the priority reasoning:
- Today's fires come first. Lead magnet 500 bug is blocking lead capture in production. Stitch needs to update the Railway env var, not build plugins.
- Scorecard needs to ship. Migration 021, deploy, test — this is the next revenue-relevant milestone.
- Kit trial expires March 5. We need the email flows working and tested before we're paying $29+/month.
- The current process works. Our CLAUDE.md checklists and inbox system aren't broken — they're just manual. "Manual but working" beats "automated but half-built" at this stage.
When to revisit: After scorecard is live, lead magnet is fixed, and email flows are QA'd. Probably late this week or early next week. At that point, I'd build session-open and session-close first (biggest ROI), then expert-panel if there's time.
One uncertainty: I haven't tested whether custom skills built on one machine (the Mini) would transfer if we ever change hardware or if Pixel on the HP could use Charlotte-built skills. The docs say plugins are "stored locally" with org-wide sharing "coming soon."