Hugging Face Hub MCP Server

Instructions

Implementation Reference

• src/tools/datasets.ts:251-275 (handler)

  The handler function that executes the tool logic: validates input, calls the HuggingFaceClient to fetch Croissant metadata, and returns the result or error.

  export async function handleGetCroissant(client: HuggingFaceClient, args: unknown): Promise<CallToolResult> {
      try {
          if (!isCroissantArgs(args)) {
              throw new Error("Invalid arguments for hf_get_croissant");
          }
  
          const { repo_id } = args;
          const results = await client.getDatasetCroissant(repo_id);
          
          return {
              content: [{ type: "text", text: results }],
              isError: false,
          };
      } catch (error) {
          return {
              content: [
                  {
                      type: "text",
                      text: `Error: ${error instanceof Error ? error.message : String(error)}`,
                  },
              ],
              isError: true,
          };
      }
  }

• src/tools/datasets.ts:114-127 (schema)

  The tool definition including name, description, and input schema (requires repo_id).

  export const getCroissantToolDefinition: Tool = {
      name: "hf_get_croissant",
      description: "Get the Croissant metadata for a dataset. Croissant is a high-level format for machine learning datasets.",
      inputSchema: {
          type: "object",
          properties: {
              repo_id: {
                  type: "string", 
                  description: "Dataset repository ID"
              }
          },
          required: ["repo_id"]
      }
  };

• src/server.ts:55-66 (registration)

  Registration of the tool definition in the MCP server's list tools request handler.

  this.server.setRequestHandler(ListToolsRequestSchema, async () => ({
      tools: [
          listModelsToolDefinition,
          getModelInfoToolDefinition,
          getModelTagsToolDefinition,
          listDatasetsToolDefinition,
          getDatasetInfoToolDefinition,
          getDatasetParquetToolDefinition,
          getCroissantToolDefinition,
          getDatasetTagsToolDefinition
      ],
  }));

• src/server.ts:90-91 (registration)

  Mapping of the tool name to its handler function in the MCP server's call tool request switch statement.

  case 'hf_get_croissant':
      return handleGetCroissant(this.client, args);

• src/client.ts:119-126 (helper)

  Core client method that makes the HTTP GET request to the Hugging Face API endpoint for Croissant metadata and stringifies the response.

  async getDatasetCroissant(repoId: string): Promise<string> {
      try {
          const response: AxiosResponse = await this.httpClient.get(`/api/datasets/${repoId}/croissant`);
          return JSON.stringify(response.data, null, 2);
      } catch (error) {
          throw new Error(`Failed to fetch Croissant metadata: ${error instanceof Error ? error.message : String(error)}`);
      }
  }