MCP server for AI agents

Hermeticor runs a Model Context Protocol server at $BASE/mcp. An agent connects with an API key and sees only the tools that key's scopes allow.

What you need first

• An MCP client (an AI assistant or agent framework that speaks streamable HTTP).

• A secret key with mcp:use plus the scopes for what the agent may touch. For a read-only analyst, mcp:use with analytics:read and profiles:read is enough.

Give agents their own key. Never reuse your main server key.

1. Point your client at it

Most clients need the URL and a header:

{
  "mcpServers": {
    "hermetica": {
      "url": "https://hermeticor.com/mcp",
      "headers": { "Authorization": "Bearer hm_sk_..." }
    }
  }
}

2. Or call it directly

curl -s -X POST "$BASE/mcp" \
  -H "Authorization: Bearer $HERMETICA_API_KEY" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"my-agent","version":"1"}}}'

curl -s -X POST "$BASE/mcp" \
  -H "Authorization: Bearer $HERMETICA_API_KEY" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

curl -s -X POST "$BASE/mcp" \
  -H "Authorization: Bearer $HERMETICA_API_KEY" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"get_billing_usage","arguments":{}}}'

How it behaves

• Stateless: every call is a POST. A GET answers 405. A notification answers 202.

• Tools outside the key's scopes are hidden from tools/list.

• Tool calls go through the same checks as the API. An agent cannot send to someone without consent or who is suppressed.