Any MCP-capable program can connect to Claix. Common options in production and development:

• Cursor
• Claude Desktop
• Claude Code
• Windsurf
• Cline
• Continue
• Zed
• Smithery
• Lovable
• Replit Agent
• ChatGPT (with MCP connectors)
• n8n
• Make
• Zapier (via HTTP/MCP)
• LangGraph / LangChain
• Any MCP client with Streamable HTTP or SSE

IDEs and coding agents (Cursor, Windsurf, Cline…) typically use JSON config with the server URL and headers. Smithery and Streamable HTTP clients send JSON-RPC directly to POST /mcp. Claude Desktop usually uses SSE via mcp-remote (GET /mcp + POST /mcp/message).

Recommended — HTTP header on every request:

x-api-key: <YOUR_API_KEY>

In Smithery, configure the header with x-from: { "header": "x-api-key" } so users enter their key when connecting. The API key is optional in the config schema when the client already sends x-api-key.

Per-tool fallback: many tools accept api_key in arguments when the MCP client cannot send custom headers.

Both modes are active on the same base URL. The server auto-detects the request type.

Static metadata for scanning (Smithery, clients): https://claix.dev/.well-known/mcp/server-card.json

1. Call claix.schemas.list to get schema_id, type, and whether Agent mode is enabled.
2. Use claix.extract.* for structured extraction or claix.agent.* when is_agent_mode=true.
3. Send the file as file_base64 (Base64 or data URL). The server rebuilds multipart requests to the Claix REST API.
4. Read the response in structuredContent.data (typed JSON) or the text in content.
5. If the schema has context window enabled and extraction returns a document_id, keep that UUID — it references the document persisted in Claix memory.
6. With document_id, use claix.window_context.get to read raw content, claix.window_context.ask for natural-language questions, or claix.document.delete to remove the document when you no longer need it.

When a schema has context window enabled, Claix persists the processed document and returns a document_id in the extraction response. The three tools below are direct proxies to the context window and delete REST APIs; they do not consume extraction credits.

Common errors in structuredContent.error: 401 (invalid API key), 404 (document not found or not owned by your account), 400 (invalid parameters). Delete accepts only DELETE with no body.

Related workflow prompt: workflow.window-context-ask (via prompts/get).

After connecting, call tools/list for all 23 tools with inputSchema, outputSchema, and annotations (including destructiveHint on claix.document.delete):

The server exposes reusable prompts via prompts/list:

Documentation resources available via resources/list:

One-click install:

npx -y @smithery/cli run info-f4xz/claix

Manual configuration:

{
  "mcpUrl": "https://claix.dev/mcp",
  "headers": {
    "x-api-key": "YOUR_API_KEY"
  }
}

Directory listing: smithery.ai/server/info-f4xz/claix

Add to Cursor MCP settings (Settings → MCP):

{
  "mcpServers": {
    "claix": {
      "url": "https://claix.dev/mcp",
      "headers": {
        "x-api-key": "YOUR_API_KEY"
      }
    }
  }
}

{
  "mcpServers": {
    "claix": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://claix.dev/mcp",
        "--header",
        "x-api-key:YOUR_API_KEY"
      ]
    }
  }
}

In n8n, use an HTTP Request node with POST https://claix.dev/mcp, JSON-RPC headers, and body {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"claix.extract.pdf","arguments":{"schema_id":"…","file_base64":"…"}}}. For context window, set name to claix.window_context.get, claix.window_context.ask, or claix.document.delete with document_id in arguments. Community MCP nodes also work — point the base URL to Claix and pass x-api-key.