Syncplify.me Server v8

Appearance

Automation and AI agents ​

Syncplify Server! is managed through one REST API, the same one every console of the product is built on. An orchestration tool or an AI agent that talks to the server needs three things from that API: a way to observe the server, a safe way to act on it, and, later, a way to be woken up by it. This chapter documents what the server offers for all three: identity and safe writes from release 8.8.0, outbound events from release 8.9.0, and the MCP adapter that puts the whole management API in front of an AI agent from release 8.10.0.

  • Automation tokens: unattended credentials for the SuperAdmin and Admin APIs, bound to the operator who minted them and never broader than that operator.
  • Safe writes: dry runs, idempotent creates and optimistic concurrency, so that a caller that guesses, retries or races a human never does damage.
  • The served specification: the server describes its own API at a stable URL, with its real version and address.
  • Reason codes: the stable, machine readable code every refusal carries.
  • Webhooks: the server calls your receiver with signed JSON for the log events you subscribe to, with retries and a delivery ID, from every node. Release 8.9.0.
  • The live event stream: the same envelopes over server sent events, filtered by event, block and level. Release 8.9.0.
  • The MCP adapter: ss-mcp, a stand alone Model Context Protocol server that exposes the management API as tools to Claude Desktop, Claude Code and other agents, with exactly the rights of its automation token and a preview and confirm step on every write. Where to download it, how to install it and how to connect each client. Release 8.10.0.
  • Working with an AI agent: how a conversation with the connected agent goes, what to ask it, what it refuses, and how to keep it safe, with worked examples.

Everything in this chapter is off until a SuperAdmin turns it on, under Global Configuration, Automation. It needs a Professional or Ultimate license.

The security model in one paragraph ​

A token is not an account. It belongs to the operator who created it, it can do at most what that operator can do at the moment of the request, and it stops working when that operator is disabled or deleted. It is read only unless the operator says otherwise, it can never touch another credential (passwords, second factors, passkeys, operator accounts, backups), and it needs a separate scope to change the service itself (restart or drain a node, start or stop a site, update the software, renew a certificate, terminate sessions). Every change made with a token is recorded with both the token and its owner. One switch disables every token at once.

What is deliberately not here ​

There is no language model inside the product, no natural language configuration, and no listener for remote agent protocols. The agent is yours; the server is the tool it uses. The MCP adapter runs next to the agent and speaks to the server over the same REST API as every console; the server itself hosts no agent protocol.