Testing MCP with Claude Code running locally, main benefit right now seems to be that the combo automates some parts of my development workflow, e.g. no more manually pulling named branches from browser-based sessions, then copy and pasting error messages as I debug getting new features working locally. Local access to the live app also appears useful even if I'm writing features by hand, plus read-only database access means some tasks, like checking that seed data actually landed, are easier? Though I probably still want the seeding itself handled reproducibly, as artifacts under version control in the application's source code. So feature work still happens in a regular Claude Code session, just with these tools along for the ride whenever something needs checking against the running app. Neat!