Interfaces
MCP for agents
Every CLI/API verb is exposed as a Model Context Protocol tool. An agent becomes a first-class operator — with the same guardrails as a human: mandatory approval and a complete audit trail. Available on every tier.
Why MCP
MCP lets an agent (yours, or our internal automation) run the entire loop: discover accounts → draft → reformat per network → request approval → schedule → analyze → iterate. Unlike tools that bolt on a chatbot, Socializer makes the agent a real operator — but one that cannot publish to a connected account without a human approving the draft first.
Exposed tools
The MCP server surfaces the full verb set, including:
compose,reformat,thread— drafting & per-network adaptationapprove— request/record the human approval gateschedule,post_now,queue_list— publishing & queue controlprofiles_status,analytics,mentions_list,mentions_reply— status & engagement
Configure the MCP server
Point your MCP client at the Socializer server and authenticate with a tenant API key. In a Claude Code / generic MCP client config:
{
"mcpServers": {
"socializer": {
"command": "socializer",
"args": ["mcp"],
"env": {
"SOCIALIZER_API_URL": "https://socializer.co",
"SOCIALIZER_API_KEY": "sk_live_…"
}
}
}
}The server authenticates with the same tenant-scoped key as the CLI and API, so the agent is bound to exactly one workspace and its row-level-security isolation.
Guardrails are non-negotiable
- Approval gates — any write to a client/brand account requires human sign-off. Autonomous posting is off by default and limited to explicitly-flagged owned channels.
- Immutable audit log — every agent action records actor=agent, the content, the target, and the result. Full provenance and reversibility.
- Dry-run — agents can preview the exact payload without publishing.
The buyer pays for control over the AI, not abdication to it. An agent that posts off-brand content to a client account is a contract-losing event — Socializer is designed so that cannot happen without a human in the loop.