ElevenAgents by ElevenLabsScale conversations without scaling your team
Promoted
Maker
📌
Qore is a single-binary TUI that manages 8 infrastructure services + exposes everything through an MCP (Model Context Protocol) server — so you can control your infra both manually via keyboard and through AI assistants like Claude or Cursor.
How the tools work:
Each connection type (Redis, Postgres, MySQL, Mongo, S3, HTTP, SSH, Git) has a native protocol driver — no CLI wrappers, no dependencies. Redis speaks RESP directly via Bun.connect, S3 signs requests with AWS SigV4, SSH uses ssh2 with full PTY support.
The TUI is data-driven: commands are arrays, screens are components, everything scrolls with block-based rendering. Type a command, get results inline.
Multi-tab: all connections stay mounted simultaneously. Switch with Ctrl+Tab, state preserved.
The MCP server is the key differentiator:
It exposes 35 tools (SSH exec, Docker container ops, database queries, system discovery, HTTP requests) over JSON-RPC 2.0 via a local TCP socket.
5 resources: connections list, probe snapshots, Docker containers/images, system info.
4 prompts: infrastructure diagnostics, security audit, container health, database health.
Zero credential exposure: AI tools reference connections by name only. The vault bridge resolves credentials server-side — the AI never sees passwords or SSH keys.
The vault uses AES-256-GCM + scrypt. Master password never hits disk. A Unix socket bridge (~/.qore/qore.sock, chmod 0600) connects the vault to MCP without exposing secrets.
Install: curl -fsSL https://github.com/Kodjaoglanian... | bash
Code: https://github.com/Kodjaoglanian...
Report
How does the MCP server piece work day to day, is it running as a sidecar process or baked into the TUI itself? Trying to figure out if there are extra steps to wire it into something like Claude or Cursor.
Report
How does the MCP server actually expose the terminal context to external AI clients, and is there any sandboxing in place so a misbehaving model cant accidentally rm something it shouldnt?
How does the MCP server piece work day to day, is it running as a sidecar process or baked into the TUI itself? Trying to figure out if there are extra steps to wire it into something like Claude or Cursor.
How does the MCP server actually expose the terminal context to external AI clients, and is there any sandboxing in place so a misbehaving model cant accidentally rm something it shouldnt?