Dumpling AI MCP Server

Instructions

Implementation Reference

• src/index.ts:666-687 (handler)

  The handler function for the 'extract-document' tool. It proxies the request to the external Dumpling AI API endpoint /api/v1/extract-document, handles authentication with API key, and returns the response as MCP content.

      const apiKey = process.env.DUMPLING_API_KEY;
      if (!apiKey) throw new Error("DUMPLING_API_KEY not set");
      const response = await fetch(`${NWS_API_BASE}/api/v1/extract-document`, {
        method: "POST",
        headers: {
          "Content-Type": "application/json",
          Authorization: `Bearer ${apiKey}`,
        },
        body: JSON.stringify({
          inputMethod,
          files,
          prompt,
          jsonMode,
          requestSource: "mcp",
        }),
      });
      if (!response.ok)
        throw new Error(`Failed: ${response.status} ${await response.text()}`);
      const data = await response.json();
      return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
    }
  );

• src/index.ts:658-665 (schema)

  Zod schema defining the input parameters for the 'extract-document' tool: inputMethod (url or base64), files (array of strings), prompt (string), jsonMode (optional boolean).

    inputMethod: z.enum(["url", "base64"]).describe("Input method"),
    files: z
      .array(z.string())
      .describe("Array of URLs or base64-encoded documents"),
    prompt: z.string().describe("Extraction prompt"),
    jsonMode: z.boolean().optional().describe("Return in JSON format"),
  },
  async ({ inputMethod, files, prompt, jsonMode }) => {

• src/index.ts:655-688 (registration)

  Registration of the 'extract-document' tool using McpServer.tool(), including name, description, input schema, and handler function.

    "extract-document",
    "Extract structured data from documents based on a prompt.",
    {
      inputMethod: z.enum(["url", "base64"]).describe("Input method"),
      files: z
        .array(z.string())
        .describe("Array of URLs or base64-encoded documents"),
      prompt: z.string().describe("Extraction prompt"),
      jsonMode: z.boolean().optional().describe("Return in JSON format"),
    },
    async ({ inputMethod, files, prompt, jsonMode }) => {
      const apiKey = process.env.DUMPLING_API_KEY;
      if (!apiKey) throw new Error("DUMPLING_API_KEY not set");
      const response = await fetch(`${NWS_API_BASE}/api/v1/extract-document`, {
        method: "POST",
        headers: {
          "Content-Type": "application/json",
          Authorization: `Bearer ${apiKey}`,
        },
        body: JSON.stringify({
          inputMethod,
          files,
          prompt,
          jsonMode,
          requestSource: "mcp",
        }),
      });
      if (!response.ok)
        throw new Error(`Failed: ${response.status} ${await response.text()}`);
      const data = await response.json();
      return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
    }
  );