Agent Switch's product goal is one control plane for MCP definitions and static credential values across local AI agents. The project documents incomplete boundaries explicitly instead of treating a successful demo as a finished distribution.
- Neutral first run with no maintainer-specific MCPs or credentials.
- Central lifecycle for user-level command/stdio MCPs.
- Explicit import preview and backed-up adoption from supported agents.
- One private credential store with per-MCP least-privilege wrappers.
- Detected-target-only reconciliation, strict health checks, CI, and package build validation.
- Normalize native Streamable HTTP and SSE MCP definitions without losing client-specific OAuth behavior; migrate static header credentials without printing or duplicating their values.
- Discover project/local MCP scopes separately from user scope and require an explicit ownership choice before adoption.
- Add adapters through a documented contract and fixture suite rather than guessing unknown agent configuration formats.
- Replace the current ad-hoc signed Homebrew alpha with a Developer ID-signed and notarized release, then remove the Cask's Gatekeeper-quarantine bypass.
- Add complete English and Chinese localization plus a first-run wizard that previews targets, MCP IDs, and secret names before any write.
- Add a first-class backup browser and guided restore flow.