What is MCP & Setup

MCP (Model Context Protocol) is an open standard that lets AI assistants — like Claude — call external systems as a set of tools. Synaptyx ships an MCP server, so you can talk to your tracking data: ask for an overview, find anomalies, or (with confirmation) make changes like pausing a campaign.

What you can do with it

• Ask questions: “How did my French campaigns do this week?” → the assistant calls get_overview, get_breakdown, etc.
• Get analysis: anomalies, optimization suggestions, funnel bottlenecks.
• Take action (safely): update a flow, create a routing rule, set a budget, create an alert, blacklist a source — all dry-run by default.
• Drive Facebook ads: read campaigns/ad sets/insights and (with explicit confirmation) create or scale them.

See the full Tool Catalogue.

Authentication & plan gate

The MCP server is at POST /api/mcp. It authenticates with an API key that has the mcp:* scope, and that scope requires a Pro or Scale plan.

Because scopes are re-checked against your current plan on every request, a key minted on Pro that later downgrades will get a clear 403 rather than silently working.

Setting up a client

npx -y mcp-remote https://app.synaptyx.pro/api/mcp \
  --header "Authorization: ApiKey syx_live_your_key_here"

{
  "mcpServers": {
    "synaptyx": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://app.synaptyx.pro/api/mcp",
        "--header", "Authorization: ApiKey syx_live_your_key_here"
      ]
    }
  }
}

How it works under the hood

• Transport: Streamable HTTP / JSON-RPC 2.0 at POST /api/mcp (protocol version 2024-11-05, server name synaptyx).
• Surface: the server advertises tools only (no resources/prompts).
• Standard methods: initialize, tools/list, tools/call, ping.

Safety, limits & audit

Next