m2ai-portfolio MCP Portability Auditor
Classifies every extension, plugin, connector, and skill in a user's agent stack as either Open (MCP-compatible) or Proprietary (vendor-locked), and produces a portability score.
When to Invoke
Trigger on: "audit my MCP setup", "which of my plugins are portable", "check my extensions for lock-in", "score my agent stack for portability", "can I move my GPTs to Claude", "is this skill MCP-compatible".
Inputs
Ask the user (skip any already answered):
- Claude Code plugins / skills directory — usually
~/.claude/plugins/and~/.claude/skills/ - MCP servers configured — check
~/.claude.jsonor Claude Desktop config - ChatGPT / Custom GPTs in active use — names and whether they rely on proprietary actions
- Gemini extensions / Gems — any active extensions
- Conway
.cnw.zipfiles — if any (speculative until Conway ships) - Cursor/Copilot/other IDE extensions — if relevant
Classification Rules
For each item, classify as:
- OPEN (MCP) — implemented via Model Context Protocol, works across any MCP-speaking client. Portability score: 5/5.
- OPEN-ADJACENT — uses a standard API (REST, OAuth) but ties to a single client's UI/invocation. Portable with moderate rework. Score: 3/5.
- PROPRIETARY — vendor-only format (
.cnw.zip, Custom GPT actions, Gemini Gem manifests, Claude skills without MCP equivalents). Score: 1/5. - LOCKED — contains proprietary state (training data, tuned behavior) that cannot be exported at all. Score: 0/5.
Output Format
# MCP Portability Audit — {date}
## Inventory
| Tool | Host | Format | Category | Score |
|------|------|--------|----------|-------|
| ... | Claude Code | MCP server | OPEN | 5/5 |
## Portability Score
Overall: {weighted avg}/5 ({count} tools audited)
## Breakdown by Host
- Claude Code: N tools, avg X/5
- ChatGPT: ...
- Gemini: ...
## Migration Candidates (high value, currently locked)
1. {tool} — {why it matters} — {MCP equivalent or rebuild cost}
## Keep As-Is
Tools that are either already portable or whose lock-in cost is justified.
## Open-Source MCP Equivalents Worth Adopting
{list of community MCP servers that would replace proprietary items}
Verification
- Every item has an explicit host, format, and category
- Proprietary items include a concrete migration path (or "no MCP equivalent yet — accept lock-in")
- Claude skills that are pure prompt content are flagged as PORTABLE regardless of host
- Output saved to
/tmp/mcp-portability-audit-{date}.mdunless user specifies otherwise
Source
Nate's Newsletter — "512,000 Lines of Leaked Code Reveal the Lock-In Strategy Coming for Your AI Stack" (2026-04-08). Builds on Nate's observation that Conway's .cnw.zip format is an explicit bet against MCP as the portability standard.