Orion Vision MCP Server

Instructions

Implementation Reference

• src/tools/form-recognizer.ts:11-35 (handler)

  The core handler function that downloads the document from the URL, analyzes it using Azure Form Recognizer client, and returns the JSON string result.

  export const analyzeDocument = async (url: string, modelId?: string): Promise<string> => {
    const config = getConfig();
    const client = new DocumentAnalysisClient(
      config.formRecognizerEndpoint,
      new AzureKeyCredential(config.formRecognizerKey)
    );
  
    try {
      // Download the document from URL
      const response = await axios.get(url, { responseType: "arraybuffer" });
      const buffer = Buffer.from(response.data);
  
      // Analyze the document
      const poller = await client.beginAnalyzeDocument(
        modelId || "prebuilt-document",
        buffer as FormRecognizerRequestBody
      );
  
      const result = await poller.pollUntilDone();
      return JSON.stringify(result, null, 2);
    } catch (error) {
      const errorMessage = error instanceof Error ? error.message : 'Unknown error';
      throw new Error(`Document analysis failed: ${errorMessage}`);
    }
  };

• src/types/tools.ts:15-18 (schema)

  Zod schema defining the input parameters for the analyze-document tool: url (required string URL) and optional modelId.

  export const DocumentSchema = z.object({
    url: z.string().url().describe("URL of the document to analyze"),
    modelId: z.string().optional().describe("Optional model ID for custom models"),
  });

• src/services/tools.ts:8-15 (registration)

  Registration of the 'analyze-document' tool in the tools array, specifying name, description, schema, and execute wrapper that calls the handler.

  {
    name: "analyze-document",
    description: "Analyzes a document using Azure Form Recognizer and returns structured data",
    parameters: DocumentSchema,
    execute: async (args) => {
      return await analyzeDocument(args.url, args.modelId);
    },
  },