Every endpoint, one MCP server.

Connect a client

The server speaks the Streamable HTTP transport and is stateless — no session to manage. Authentication is the same Authorization header the HTTP API takes, with the same scopes: a key that can call /v1 can call MCP. Any client not listed takes the same shape: the URL, and the header.

claude mcp add --transport http diraz https://mcp.diraz.ae/mcp \
  --header "Authorization: Bearer drz_live_..."

Four tools, not 289

An agent's context cannot hold a tool definition per endpoint. The server exposes the same workflow a developer uses: search, read, call.

list_categories

What data exists — every category and how many endpoints each holds. Start here when exploring.

search_apis

Find endpoints by keyword: "gold price", "earthquake", "BPOM". Returns paths, parameters, and a one-line summary for each hit.

get_api_docs

One endpoint in full — parameter types and examples, response fields, rate limit, credit cost. Accepts a slug or a /v1 path.

call_api

Execute it. Path parameters fill by name; everything else goes on the query string. The JSON response comes back as tool content.

{
  "tools": [
    { "name": "list_categories", "description": "Every category and its endpoint count." },
    { "name": "search_apis", "description": "Find endpoints by keyword, vendor, or category." },
    { "name": "get_api_docs", "description": "One endpoint's parameters, fields, cost, and limits." },
    { "name": "call_api", "description": "Call the endpoint and return its JSON response." }
  ]
}

{
  "name": "call_api",
  "arguments": {
    "endpoint": "gold-spot",
    "params": {}
  }
}

Billing is unchanged

A call_api runs through the exact same pipeline as a direct HTTP request — scopes, rate limits, credits, usage recording, circuit breakers. A call made by an agent costs the same credits as one made by your code, shows up in the same usage dashboard, and fails free, like every error here. The three discovery tools cost nothing.