One MCP server,
every AI client.

What it is

Zapi runs a remote MCP server named zapi (version 1.2.0) at https://mcp.zpi.web.id/mcp. Transport is Streamable HTTP, stateless — POST carries JSON-RPC, GET opens SSE, DELETE tears down. There are no resources and no prompts: exactly 8 tools, documented in full below.

MCP (Model Context Protocol) is how AI clients — Claude, Cursor, VS Code Copilot, Gemini, Windsurf, and others — call external tools. Once connected, your agent can discover a scraper and run it without you writing any HTTP code.

Authentication

Authenticate with your Zapi API key (zpi_...) via any of three headers: Authorization: Bearer, x-api-key, or api-key. Missing or invalid keys get a 401. Create a key at Dashboard → API Keys — free, no credit card required.

# Zapi MCP endpoint (Streamable HTTP)
https://mcp.zpi.web.id/mcp

# Auth via header — any of these three:
Authorization: Bearer zpi_xxx
x-api-key: zpi_xxx
api-key: zpi_xxx

Per-client setup

Find your client, copy its config, and replace zpi_xxx with your key. The endpoint is already baked into every snippet. Clients differ in the config key they expect for a remote HTTP server — the wrong key fails silently:

claude mcp add --transport http zapi https://mcp.zpi.web.id/mcp \
  --header "Authorization: Bearer zpi_xxx"

{
  "mcpServers": {
    "zapi": {
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "httpUrl": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "serverUrl": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "servers": {
    "zapi": {
      "type": "http",
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "serverUrl": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": [
    {
      "name": "zapi",
      "type": "streamable-http",
      "url": "https://mcp.zpi.web.id/mcp",
      "requestOptions": {
        "headers": {
          "Authorization": "Bearer zpi_xxx"
        }
      }
    }
  ]
}

{
  "context_servers": {
    "zapi": {
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "url": "https://mcp.zpi.web.id/mcp",
      "headers": {
        "Authorization": "Bearer zpi_xxx"
      }
    }
  }
}

{
  "mcpServers": {
    "zapi": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.zpi.web.id/mcp",
        "--header",
        "Authorization: Bearer zpi_xxx"
      ]
    }
  }
}

SDK client

From TypeScript, skip the client config entirely — zpi-sdk/mcp is a zero-dependency MCP client for the same server: JSON-RPC over Streamable HTTP, protocol version 2025-06-18, session tracked via the mcp-session-id header. The initialize handshake runs lazily on first call.

import { createMcpClient } from "zpi-sdk/mcp";

const mcp = createMcpClient({
  apiKey: "zpi_...",
  baseURL: "https://api.zpi.web.id",
});
const tools = await mcp.listTools();
const result = await mcp.callTool("run_scraper", {
  scraper: "instagram-scraper",
  endpoint: "profile",
  params: { username: "zaadevofc" },
});

listTools() returns the 8 tool definitions; callTool(name, args) invokes one. RPC failures throw ZpiMcpError (with code and data). See the SDK docs for the full client reference.

Tools reference

All 8 tools, in registration order. Read-only tools are free; run_scraper bills per call and bulk_submit bills per item. The recommended workflow:

# The recommended 3-step workflow for your agent:
search_scrapers query="instagram profile"      # 1) find an endpoint
get_scraper_schema scraper="..." endpoint="..." # 2) only if flagged needsSchemaCall
run_scraper scraper="..." endpoint="profile" params={ "username": "zaadevofc" }

Rate limits

MCP shares rate limits with the REST API. The per-minute window counts all requests; the per-month window counts only billable ones. Free tier: 100 requests/minute, 2,000 billable requests/month — paid tiers on Pricing. A 429 response is JSON with window: "minute" | "month" telling you which limit you hit.

Plan gating

• Every endpoint declares a minPlan — calls below it fail with plan_upgrade_required (including requiredPlan and upgradeUrl).
• Bulk needs three things at once: a Pro+ plan, bulkEnabled on the endpoint, and staying under your tier's per-job item cap.
• An expired paid plan silently drops you to free limits — check get_account (planStatus) if calls start rate-limiting unexpectedly.

Error codes

Tool failures set isError with a machine code in structuredContent.code:

Troubleshooting

• 401 / unauthenticated — wrong, revoked, or missing key. Check the header name (any of Authorization: Bearer, x-api-key, api-key) and rotate at /dashboard/keys.
• Connected but unauthorized — your client is OAuth-only and never sent the header. Use the mcp-remote bridge.
• 403 from a browser — the origin guard blocks browser origins that aren't allow-listed. Call the server from a client or backend instead of in-page fetch.
• Client doesn't see the server — usually the wrong config key. Check the config-key table (url / httpUrl / serverUrl / servers).
• 429 — check which window you hit (window: "minute" | "month") and the X-RateLimit headers; a minute limit clears itself, a month limit means upgrading.
• plan_upgrade_required — the endpoint's minPlan is above your tier. See Pricing.