35 lines (23 loc) · 1.53 KB

MCP Runtime Matrix

This matrix defines the authoritative state of MCP integrations across the AI Product OS.

Integration	Codex	Claude	Cursor	VS Code	Scope
Linear	Read	Full	Read	View	Project Sync
Neon	Full	Full	Full	Full	Database Ops
Vercel	-	Full	-	View	Deployments
Codebase-Memory	Full	Full	Full	Full	Search & Graph
Playwright	-	Full	-	-	E2E Testing

Integration Guidelines

1. Codex (OpenAI)

Primary Mode: Read-only audit and planning.
Constraints: Use root-level .codex/config.toml to restrict MCP tool write-access. Focus on structural integrity and best practices.

2. Claude (Anthropic)

Primary Mode: Full Pipeline Execution.
Protocol: Claude is the canonical orchestrator. All quality gates and sequential pipeline commands are optimized for Claude Code.

3. Cursor / VS Code

Primary Mode: IDE-assisted implementation.
Protocol: Use IDE context rules (.cursor/rules or .vscode/settings.json) to enforce local coding standards while remaining within the pipeline stage.

Safety & Bounded Subagents

Bounded Exploration: High-volume tasks (e.g., massive file reads, exhaustive searches) MUST be delegated to subagents.
Audit Logging: All MCP write operations should be logged in project-state.md Decisions Log.