Purpose
AI tools can act on our systems through a known, controlled set of connectors.
The standard
Server | Gives Claude | Where configured | Owner |
|---|---|---|---|
Rhino MCP | create/inspect geometry in the running Rhino session |
|
|
Supabase MCP | schema inspection, SQL against dev projects only | per-repo | |
GitHub MCP | issues, PRs | per-repo | |
Google Drive / Gmail / Calendar | reference docs, scheduling | claude.ai connectors | individual |
MCP configs are committed per repo (
.mcp.json) with no secrets inside; tokens come from env.Any MCP with write access to production data is prohibited unless approved via ADR.
Rhino MCP: run only against a scratch document unless the tool is explicitly for a project file; save before every agent run.
Procedure — adding a server
Open an intake issue: what it exposes, read/write scope, auth model, which repos. On approval add to this table and the template .mcp.json.
Owner: Matt · Last reviewed: 2026-09