Feature: Document MCP-to-skill bindings — skill-level dependency mapping

[zaziszaz]

Problem

Users cannot discover which MCP connectors enhance a skill or enable a plugin's full feature set. This creates friction for:

• Users who want to know: 'Does comps-analysis work without FactSet? What's the difference?'
• Teams configuring which MCPs to enable for which workflows
• Skill authors who want to document what data sources augment their skill

Currently, MCP dependencies are scattered across prose in SKILL.md instructions and plugin READMEs, making them non-discoverable.

Key principle: Skills are MCP-agnostic (work standalone), but MCPs augment output quality and capability. MCPs should be documented at both levels:

• Skill level: Which MCPs enhance this skill (optional, improves output)
• Plugin level: Which MCPs unlock the plugin's full feature set (required/recommended)

Current State

financial-analysis plugin:

• .mcp.json lists 11 MCP servers available to the vertical
• README lists all MCPs (lines 117–136) but never maps them to skills or plugin value
• CLAUDE.md (line 42) references mcp-categories.json: Canonical MCP category definitions shared across plugins — file does not exist in repo
• Individual skills (e.g., comps-analysis/SKILL.md line 28) mention MCPs only in prose: "If S&P Kensho MCP, FactSet MCP, or Daloopa MCP are available, use them exclusively"
• No way to know upfront: 'Will this skill work without these MCPs? Will it be degraded?'

Comparison to knowledge-work-plugins repo:

• partner-built/zoom-plugin/CONNECTORS.md documents which MCPs exist
• Main README maps plugins to connectors (e.g., "sales plugin → Slack, HubSpot, Close, Clay, ZoomInfo, Notion, Jira, Fireflies, Microsoft 365")
• Still missing: skill-level augmentation mapping (e.g., "call-prep works standalone, but HubSpot + Slack enhance it")

Proposed Solution

Option 1: SKILL.md Frontmatter Declaration (Recommended)

Add optional enhances_with_mcp field to each SKILL.md frontmatter:

---
name: comps-analysis
description: Build institutional-grade comparable company analyses...
enhances_with_mcp:
  - factset      # Core data source; output quality 80% without it
  - sp-global    # Alternative data source; improves breadth
  - daloopa      # Secondary research layer
optional_mcp:
  - morningstar  # Nice-to-have for comparative metrics
---

Users see: 'This skill works without MCPs, but FactSet/S&P Global significantly enhance it.'

Option 2: Plugin-level CONNECTORS.md

Create/enhance plugins/vertical-plugins/<vertical>/CONNECTORS.md that documents plugin value by MCP:

# Connectors — Financial Analysis

This plugin works in **standalone mode** (reference docs only) and gets **supercharged with MCPs**.

## Core Feature Set (Standalone)

- Comps-analysis templates and structural guidance
- DCF model frameworks
- LBO calculation shells

## Supercharged Features (with MCPs)

| MCP | Unlocks | Required For |
|-----|---------|--------------|
| FactSet | Live market data, comparable multiples | comps-analysis, earnings-reviewer (output quality) |
| S&P Global | Company financials, industry benchmarks | comps-analysis, dcf-model (best results) |
| Daloopa | Research synthesis, alternative data | earnings-reviewer (depth) |
| Morningstar | Peer fund analysis | equity-research (optional enhancement) |

Option 3: Populate mcp-categories.json

Create the promised mcp-categories.json file with schema for organizing MCPs by category, skill affinity, and value tier (core/recommended/optional).

Impact

• ✅ Skills remain MCP-agnostic; MCPs are clearly optional augmentation
• ✅ Users see upfront what MCPs enhance each skill and by how much
• ✅ Teams can make informed decisions: 'Do we enable FactSet for comps-analysis?'
• ✅ Reduces friction for plugin customization and team onboarding
• ✅ Aligns financial-services documentation with knowledge-work-plugins pattern

Next Steps

1. Validate preferred approach (Option 1 + Option 2 recommended: skill-level enhancement field + plugin-level value matrix)
2. Implement for financial-analysis vertical (core skills)
3. Propagate pattern to other verticals (investment-banking, equity-research, private-equity, wealth-management, fund-admin, operations)
4. If Option 3: create mcp-categories.json schema and populate it

---

Note: This also resolves the ghost reference to mcp-categories.json in CLAUDE.md line 42 — that file should either be created or the reference removed.