One catalog. Every agent. Shared auth, policies, and tools across Cursor, Claude Code, OpenCode—or your own.
Every schema dumped into context. Tokens wasted. Poor performance.
Better, but a poor permission model that pushes users to dangerously-skip-permissions — or requires a VM.
Typed, sandboxed, shared across all agents.
Every integration indexed into one typed, discoverable catalog.
Sign in once. Every agent shares your credentials.
Auto-approve reads. Pause on writes. Wildcard rules.
Executor is an MCP server. Point any agent at it.
Secrets in your keychain. Nothing leaves your machine.
Agents search by what they need, not memorized paths.
1. Install
2. Add to your agent
You can also tell your agent to use the executor CLI directly if they don't support MCP.
3. Add a source and go
Opens the dashboard at localhost:4788 — add sources, set policies, manage auth.