SentinelX lets Claude, ChatGPT and other AI tools manage your Linux servers via the Model Context Protocol. One command on each host, OAuth via Google, and a dashboard that shows your whole fleet in one place — without configuring DNS, certificates, or an identity provider.
The hub at mcp.sentinelx.app handles OAuth, identity, and the MCP protocol. The agent on your server only opens an outbound connection to the hub — no inbound ports, no public DNS for your host, no certificates to manage.
Run the installer on your Linux server. It creates a system user, installs the agent under /opt/sentinelx-cloud-core, registers a systemd service, and connects to the hub.
Visit mcp.sentinelx.app, sign in with Google, click Enroll a new host, and paste the token the installer asks for. Your server appears in the dashboard within seconds.
The dashboard lists every enrolled server with a status dot, hostname, agent version, and last-seen time. Label hosts, kick stale sessions, or unenroll — all from the browser.
Add mcp.sentinelx.app as an MCP connector in Claude or ChatGPT. The AI sees your hosts, can target them by name, and only runs operations the agent's allowlist permits.
Every security decision in SentinelX is intentional and explicit. The agent is open source — you can audit exactly what runs on your server.
The agent only runs commands explicitly listed in its policy. Anything not in the allowlist is rejected before execution. Servers, paths and arguments are validated against the policy too.
The agent runs as a dedicated sentinelx system user with no login shell. Its sudoers rule is scoped to the exact commands the policy permits.
The hub authenticates you via Google sign-in. Each session is a JWT with explicit scopes. The agent only accepts connections from the hub, signed with your account's identity.
The agent opens an outbound WebSocket to mcp.sentinelx.app. No inbound ports, no public DNS for your host, no certificates to manage. Your firewall stays closed.
Every operation is logged on the hub with the user, host, command, output, return code, and timestamp. Blocked operations are logged too.
The agent is Apache 2.0-licensed and lives at pensados/sentinelx-cloud-core. Read it, fork it, run a fork — it's yours.
SentinelX speaks the Model Context Protocol over Streamable HTTP. Connect it to Claude, ChatGPT or any MCP-compatible agent — the same hub URL works for all of them.
You sign in once with Google. From there, every AI client you connect sees the same fleet of hosts.
https://mcp.sentinelx.app/mcpMost AI coding tools start every session from zero — a fresh context, a stale CLAUDE.md, the same decisions explained again. SentinelX runs inside your chat, so the context, the rationale, and the history accumulate across days and weeks. Combine the agent with git on the same host and you get a remote dev environment where the AI actually remembers your project.
Clone once, refactor over a week. The assistant remembers what you tried on Monday when you pick it up on Thursday — no scratchpad files to maintain, no re-explaining the constraints.
"Remember the timeout we saw last week? It came back." The chat holds the thread; the agent re-runs the diagnostics on the live host, with the previous findings already in scope.
Why you picked Postgres over SQLite. Why the worker lives on a separate host. Why that endpoint is async. The reasoning lives in the conversation, not in scattered comments you'll forget to grep for.
This site was built and is continuously updated with SentinelX.
The agent edits /var/www/sentinelx.pensa.ar/index.html directly, with git tracking every change. Including this paragraph.
If you'd rather run the entire stack yourself — including the OAuth provider — there's a Docker edition that bundles agent, MCP bridge, and Keycloak in one stack. Same MCP protocol, same security model, no managed service.
One-line install: 4 containers (agent, MCP bridge, Keycloak, Postgres) on your own host. You bring DNS, certificates, and operational responsibility. Recommended if you want full data sovereignty or air-gapped deployments.