MCP Server

Connect AI agents to Statsnet over the Model Context Protocol: client setup, tools and their arguments, limits and example prompts.

Statsnet runs a remote MCP server, so AI agents in Claude, Cursor, VS Code or any MCP-compatible client can search companies and pull their data right inside the conversation. Nothing to install locally:

POST https://statsnet.co/mcp

Access modes

Without a key (anonymous). Two tools are available — search_companies and get_company — with free public data. Anonymous search is limited to 30 calls per hour per IP and counts toward the same free monthly search quota as the website.

With an API key. Create a key in your account and pass it as Authorization: Bearer sk_live_... in the connection settings — never in the chat message itself. The key inherits its owner's subscription: search without the IP limit (up to 50 results per call) plus the paid tools. Billing matches the REST API: on API plans every tool call counts toward the monthly allowance; on report plans a full company card spends one report, and repeat views of the same company within 2 weeks are free.

An invalid or revoked key gets HTTP 401 — the server never silently downgrades such a request to anonymous mode.

Tools

Arguments

Jurisdictions: kz (Kazakhstan), uz (Uzbekistan), kg (Kyrgyzstan), ge (Georgia), am (Armenia), tj (Tajikistan), az (Azerbaijan) and US states as us-<state code> — e.g. us-de, us-ca, us-ny. The authoritative list is the inputSchema your client receives from tools/list.

Typical flow: search_companies → take the identifier from the result → get_company or a paid tool with the same jurisdiction + identifier pair. If the company is not found, no quota is spent.

Setup

Claude Code

claude mcp add --transport http statsnet https://statsnet.co/mcp \
  --header "Authorization: Bearer sk_live_YOUR_KEY"

Without --header the server connects in anonymous mode. To verify, run /mcp in Claude Code — it lists statsnet and its tools.

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json in the project:

{
  "mcpServers": {
    "statsnet": {
      "url": "https://statsnet.co/mcp",
      "headers": { "Authorization": "Bearer sk_live_YOUR_KEY" }
    }
  }
}

VS Code (GitHub Copilot agent mode)

.vscode/mcp.json:

{
  "servers": {
    "statsnet": {
      "type": "http",
      "url": "https://statsnet.co/mcp",
      "headers": { "Authorization": "Bearer sk_live_YOUR_KEY" }
    }
  }
}

Claude Desktop

Anonymous mode: Settings → Connectors → Add custom connector, URL https://statsnet.co/mcp.

With an API key: claude_desktop_config.json only runs local (stdio) servers, so connect the remote server with a header through the mcp-remote bridge (requires Node.js):

{
  "mcpServers": {
    "statsnet": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://statsnet.co/mcp",
        "--header",
        "Authorization:${STATSNET_AUTH}"
      ],
      "env": { "STATSNET_AUTH": "Bearer sk_live_YOUR_KEY" }
    }
  }
}

Restart Claude Desktop after editing the config.

Other clients

Any client that supports Streamable HTTP connects to https://statsnet.co/mcp with an optional Authorization header. Keep configs that contain a real key in personal settings, not in a shared repository.

Example prompts

One question can trigger several tool calls, and paid calls spend quota. If your client supports confirming tool calls, turn it on; when checking several companies, ask the assistant to outline its plan first.

Testing without a client (curl)

The server answers plain JSON-RPC requests — handy for debugging a key:

curl -s https://statsnet.co/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Authorization: Bearer sk_live_YOUR_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_companies","arguments":{"query":"Kazpost","jurisdiction":"kz","limit":5}}}'

The response is a JSON-RPC object with the Markdown in result.content[0].text. To see the tools available to your key: {"jsonrpc":"2.0","id":1,"method":"tools/list"}.

Errors and limits

Tool-level errors arrive as a normal result flagged isError, not as a protocol error, so the model can read the explanation and adjust. Check remaining quota in your account and plans on the pricing page.

Next steps

• MCP for business — use cases and a comparison with the REST API.
• Authentication — keys, permissions, plans.
• OpenAPI reference — the same data over the REST API.