HF Dataset MCP

Instructions

Implementation Reference

• src/tools/search-dataset.ts:44-62 (handler)

  The handler function for the 'search_dataset' tool which calls the search endpoint.

  async ({ dataset, config, split, query, offset, length }) => {
    const data = await fetchDatasetViewer<SearchResponse>("/search", {
      dataset,
      config,
      split,
      query,
      offset: offset ?? 0,
      length: length ?? 100,
    });
  
    return {
      content: [
        {
          type: "text" as const,
          text: JSON.stringify(data, null, 2),
        },
      ],
    };
  }

• src/tools/search-dataset.ts:25-43 (schema)

  Zod schema for validating the input arguments of the 'search_dataset' tool.

  {
    dataset: z.string().describe("Dataset ID (e.g., 'stanfordnlp/imdb')"),
    config: z.string().describe("Configuration name"),
    split: z.string().describe("Split name (train, test, validation)"),
    query: z.string().describe("Text to search for"),
    offset: z
      .number()
      .int()
      .min(0)
      .optional()
      .describe("Result offset for pagination (default: 0)"),
    length: z
      .number()
      .int()
      .min(1)
      .max(100)
      .optional()
      .describe("Number of results (default: 100, max: 100)"),
  },

• src/tools/search-dataset.ts:21-64 (registration)

  Registration function for the 'search_dataset' tool within the MCP server.

  export function registerSearchDataset(server: McpServer) {
    server.tool(
      "search_dataset",
      "Full-text search within a dataset split using BM25 ranking",
      {
        dataset: z.string().describe("Dataset ID (e.g., 'stanfordnlp/imdb')"),
        config: z.string().describe("Configuration name"),
        split: z.string().describe("Split name (train, test, validation)"),
        query: z.string().describe("Text to search for"),
        offset: z
          .number()
          .int()
          .min(0)
          .optional()
          .describe("Result offset for pagination (default: 0)"),
        length: z
          .number()
          .int()
          .min(1)
          .max(100)
          .optional()
          .describe("Number of results (default: 100, max: 100)"),
      },
      async ({ dataset, config, split, query, offset, length }) => {
        const data = await fetchDatasetViewer<SearchResponse>("/search", {
          dataset,
          config,
          split,
          query,
          offset: offset ?? 0,
          length: length ?? 100,
        });
  
        return {
          content: [
            {
              type: "text" as const,
              text: JSON.stringify(data, null, 2),
            },
          ],
        };
      }
    );
  }