chore(naming): distinct MCP server names — molecule-a2a (channel) vs molecule-platform (management) #38
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Problem
The agent↔agent A2A/channel MCP and the platform-management MCP (this repo,
MOLECULE_MCP_MODE=management, merged in #32dcd74cba) can both surface under an ambiguousmoleculename. When only the A2A bridge is connected to a session, the singlemoleculename falsely implies platform-admin capability is present — leading an operator/agent to reach for raw SQL / raw API instead of the OpenAPI-SSOT management surface (observed 2026-06-04: raw-SQL fallback with wrong column/table guesses because the management MCP was not wired in and the absence wasn't obvious).Convention (CTO ruling 2026-06-04)
Give the two MCPs distinct names so a session can tell at a glance which capability is present:
molecule-a2amcp__molecule_a2a__*molecule-platformmcp__molecule_platform__*Then
molecule-platformbeing absent from a session is an unambiguous signal: "no admin surface — do not fall back to raw SQL."Scope
name/ display) tomolecule-platform.moleculeplugin) rename tomolecule-a2a..mcp.jsonexamples + docs (docs repo) to the new names.molecule-platform(mode=management) into the operator/CTO Claude Code session so platform-state queries go through the OpenAPI-SSOT interface, not raw DB.Pure naming + wiring — derives no contract changes. Relates to #32 (management MCP build) and the OpenAPI SSOT (core #2056).