@llmindset/hf-mcp 0.3.11 → 0.3.13

package/src/hub-inspect.test.ts

@@ -0,0 +1,28 @@
+import { describe, expect, it } from 'vitest';
+import { HUB_REPO_DETAILS_TOOL_CONFIG } from './hub-inspect.js';
+
+describe('HUB_REPO_DETAILS_TOOL_CONFIG', () => {
+	it('defaults to overview and accepts dataset viewer operations', () => {
+		const parsed = HUB_REPO_DETAILS_TOOL_CONFIG.schema.parse({
+			repo_ids: ['rajpurkar/squad'],
+			repo_type: 'dataset',
+			operations: ['dataset_structure', 'dataset_preview'],
+			config: 'plain_text',
+			split: 'train',
+			offset: 0,
+			limit: 5,
+		});
+
+		expect(parsed.include_readme).toBe(false);
+		expect(parsed.operations).toEqual(['dataset_structure', 'dataset_preview']);
+	});
+
+	it('does not expose a redundant readme operation', () => {
+		expect(() =>
+			HUB_REPO_DETAILS_TOOL_CONFIG.schema.parse({
+				repo_ids: ['rajpurkar/squad'],
+				operations: ['readme'],
+			})
+		).toThrow();
+	});
+});

package/src/hub-inspect.ts

@@ -4,12 +4,17 @@ import { ModelDetailTool } from './model-detail.js';
 import { DatasetDetailTool } from './dataset-detail.js';
 import { spaceInfo } from '@huggingface/hub';
 import { formatDate } from './utilities.js';
+import { DatasetViewerInspector } from './dataset-viewer-inspect.js';
+
+const HUB_INSPECT_OPERATIONS = ['overview', 'dataset_structure', 'dataset_preview'] as const;
 
 export const HUB_REPO_DETAILS_TOOL_CONFIG = {
 	name: 'hub_repo_details',
 	description:
 		'Get details for one or more Hugging Face repos (model, dataset, or space). ' +
-		'Auto-detects type unless specified.',
+		'Auto-detects type unless specified. For datasets, use operations: overview, dataset_structure, dataset_preview. ' +
+		'Use dataset_structure first to discover configs, splits, sizes, and schema. Use dataset_preview only when ' +
+		'config and split are known, unless the dataset has a single config/split.',
 	schema: z.object({
 		repo_ids: z
 			.array(z.string().min(1))
@@ -18,6 +23,26 @@ export const HUB_REPO_DETAILS_TOOL_CONFIG = {
 			.describe('Repo IDs for (models|dataset/space) - usually in author/name format (e.g. openai/gpt-oss-120b)'),
 		repo_type: z.enum(['model', 'dataset', 'space']).optional().describe('Specify lookup type; otherwise auto-detects'),
 		include_readme: z.boolean().default(false).describe('Include README from the repo'),
+		operations: z
+			.array(z.enum(HUB_INSPECT_OPERATIONS))
+			.optional()
+			.describe(
+				'Details to return. Defaults to ["overview"]. For datasets, prefer ["overview", "dataset_structure"] first; then call ["dataset_preview"] with config and split.'
+			),
+		config: z
+			.string()
+			.optional()
+			.describe(
+				'Dataset Viewer config. Required for dataset_preview when the dataset has multiple config/split options. Discover via dataset_structure.'
+			),
+		split: z
+			.string()
+			.optional()
+			.describe(
+				'Dataset Viewer split. Required for dataset_preview when the dataset has multiple config/split options. Discover via dataset_structure.'
+			),
+		offset: z.number().int().nonnegative().optional().describe('Row offset for dataset_preview. Defaults to 0.'),
+		limit: z.number().int().optional().describe('Row count for dataset_preview. Defaults to 5 and is clamped to 1-100.'),
 	}),
 	annotations: {
 		title: 'Hub Repo Details',
@@ -32,11 +57,13 @@ export type HubInspectParams = z.infer<typeof HUB_REPO_DETAILS_TOOL_CONFIG.schem
 export class HubInspectTool {
 	private readonly modelDetail: ModelDetailTool;
 	private readonly datasetDetail: DatasetDetailTool;
+	private readonly datasetViewer: DatasetViewerInspector;
 	private readonly hubUrl?: string;
 
 	constructor(hfToken?: string, hubUrl?: string) {
 		this.modelDetail = new ModelDetailTool(hfToken, hubUrl);
 		this.datasetDetail = new DatasetDetailTool(hfToken, hubUrl);
+		this.datasetViewer = new DatasetViewerInspector(hfToken, { hubUrl });
 		this.hubUrl = hubUrl;
 	}
 
@@ -46,7 +73,7 @@ export class HubInspectTool {
 
 		for (const id of params.repo_ids) {
 			try {
-				const section = await this.inspectSingle(id, params.repo_type, includeReadme);
+				const section = await this.inspectSingle(id, params, includeReadme);
 				parts.push(section);
 				successCount += 1;
 			} catch (err) {
@@ -66,20 +93,30 @@ export class HubInspectTool {
 
 	private async inspectSingle(
 		repoId: string,
-		type: 'model' | 'dataset' | 'space' | undefined,
+		params: HubInspectParams,
 		includeReadme: boolean
 	): Promise<string> {
+		const type = params.repo_type;
+		const operations = normalizeOperations(params.operations);
+		const hasDatasetOperation = operations.some((operation) => operation === 'dataset_structure' || operation === 'dataset_preview');
+
 		// If caller constrained the type, do only that
 		if (type === 'model') {
+			if (hasDatasetOperation) return operationMismatch(repoId, 'model', operations);
 			return (await this.modelDetail.getDetails(repoId, includeReadme)).formatted;
 		}
 		if (type === 'dataset') {
-			return (await this.datasetDetail.getDetails(repoId, includeReadme)).formatted;
+			return await this.getDatasetDetails(repoId, params, includeReadme, operations);
 		}
 		if (type === 'space') {
+			if (hasDatasetOperation) return operationMismatch(repoId, 'space', operations);
 			return await this.getSpaceDetails(repoId);
 		}
 
+		if (hasDatasetOperation) {
+			return await this.getDatasetDetails(repoId, params, includeReadme, operations);
+		}
+
 		// Auto-detect: attempt all three and aggregate. The same id may exist for multiple types.
 		const matches: string[] = [];
 
@@ -111,6 +148,33 @@ export class HubInspectTool {
 		return matches.join('\n\n---\n\n');
 	}
 
+	private async getDatasetDetails(
+		repoId: string,
+		params: HubInspectParams,
+		includeReadme: boolean,
+		operations: HubInspectOperation[]
+	): Promise<string> {
+		const sections: string[] = [];
+		if (operations.includes('overview')) {
+			const overview = (await this.datasetDetail.getDetails(repoId, includeReadme)).formatted;
+			sections.push(`${overview}\n\n${datasetDrillDownHint()}`);
+		}
+		if (operations.includes('dataset_structure')) {
+			sections.push(await this.datasetViewer.getStructure(repoId, { config: params.config, split: params.split }));
+		}
+		if (operations.includes('dataset_preview')) {
+			sections.push(
+				await this.datasetViewer.getPreview(repoId, {
+					config: params.config,
+					split: params.split,
+					offset: params.offset,
+					limit: params.limit,
+				})
+			);
+		}
+		return sections.join('\n\n');
+	}
+
 	private async getSpaceDetails(spaceId: string): Promise<string> {
 		const additionalFields = ['author', 'tags', 'runtime', 'subdomain', 'sha'] as const;
 		const info = await spaceInfo<(typeof additionalFields)[number]>({
@@ -142,3 +206,23 @@ export class HubInspectTool {
 		return lines.join('\n');
 	}
 }
+
+type HubInspectOperation = (typeof HUB_INSPECT_OPERATIONS)[number];
+
+function normalizeOperations(operations: readonly HubInspectOperation[] | undefined): HubInspectOperation[] {
+	return operations && operations.length > 0 ? [...new Set(operations)] : ['overview'];
+}
+
+function operationMismatch(repoId: string, type: 'model' | 'space', operations: HubInspectOperation[]): string {
+	const requested = operations.filter((operation) => operation.startsWith('dataset_')).join(', ');
+	return `# ${repoId}\n\nRequested dataset operation(s) \`${requested}\`, but this repo was requested as a ${type}. Dataset Viewer operations only apply to dataset repos.`;
+}
+
+function datasetDrillDownHint(): string {
+	return [
+		'## Available deeper inspections',
+		'Call `hub_repo_details` with:',
+		'- `operations: ["dataset_structure"]` for configs, splits, sizes, parquet exports, and schema.',
+		'- `operations: ["dataset_preview"]` with `config` and `split` for sample rows.',
+	].join('\n');
+}

package/src/index.browser.ts

@@ -75,7 +75,8 @@ export const HUB_REPO_DETAILS_TOOL_CONFIG: BrowserToolConfig = {
 	name: 'hub_repo_details',
 	description:
 		'Get details for one or more Hugging Face repos (model, dataset, or space). ' +
-		'Auto-detects type unless specified.',
+		'Auto-detects type unless specified. For datasets, use dataset_structure first to discover configs, splits, ' +
+		'sizes, and schema. Use dataset_preview only when config and split are known, unless the dataset has a single config/split.',
 	annotations: {
 		title: 'Hub Repo Details',
 		destructiveHint: false,

package/src/index.ts

@@ -10,6 +10,7 @@ export * from './dataset-search.js';
 export * from './repo-search.js';
 export * from './create-repo.js';
 export * from './dataset-detail.js';
+export * from './dataset-viewer-inspect.js';
 export * from './hub-inspect.js';
 export * from './duplicate-space.js';
 export * from './space-info.js';

package/src/jobs/commands/run.ts

@@ -18,6 +18,7 @@ export async function runCommand(args: RunArgs, client: JobsApiClient, token?: s
 		secrets: args.secrets,
 		timeout: args.timeout,
 		hfToken: token,
+		volumes: args.volumes,
 	});
 
 	// Submit job
@@ -39,7 +40,11 @@ export async function runCommand(args: RunArgs, client: JobsApiClient, token?: s
 
 	// Not detached - fetch logs
 	const logsUrl = client.getLogsUrl(job.id, job.owner.name);
-	const logResult = await fetchJobLogs(logsUrl, { token, maxDuration: DEFAULT_LOG_WAIT_MS, maxLines: DEFAULT_MAX_LOG_LINES });
+	const logResult = await fetchJobLogs(logsUrl, {
+		token,
+		maxDuration: DEFAULT_LOG_WAIT_MS,
+		maxLines: DEFAULT_MAX_LOG_LINES,
+	});
 
 	let response = `Job started: ${job.id}\n\n`;
 
@@ -80,6 +85,7 @@ export async function uvCommand(args: UvArgs, client: JobsApiClient, token?: str
 		timeout: args.timeout,
 		detach: args.detach,
 		namespace: args.namespace,
+		volumes: args.volumes,
 	};
 
 	return runCommand(runArgs, client, token);

package/src/jobs/commands/scheduled.ts

@@ -28,6 +28,7 @@ export async function scheduledRunCommand(
 		secrets: args.secrets,
 		timeout: args.timeout,
 		hfToken: token,
+		volumes: args.volumes,
 	});
 
 	// Create scheduled job spec
@@ -78,6 +79,7 @@ export async function scheduledUvCommand(
 		timeout: args.timeout,
 		detach: args.detach,
 		namespace: args.namespace,
+		volumes: args.volumes,
 	};
 
 	return scheduledRunCommand(scheduledRunArgs, client, token);

package/src/jobs/commands/utils.ts

@@ -1,4 +1,4 @@
-import type { JobSpec } from '../types.js';
+import type { JobSpec, JobVolume, JobVolumeType } from '../types.js';
 import { parse as parseShellArgs } from 'shell-quote';
 
 interface EnvToken {
@@ -7,6 +7,17 @@ interface EnvToken {
 }
 
 const SPECIAL_PARAMS = new Set(['*', '@', '#', '?', '!', '-', '_']);
+const HF_VOLUME_PREFIX = 'hf://';
+const VOLUME_FORMAT_HELP =
+	'Expected format: hf://[TYPE/]OWNER/NAME[/PATH]:/MOUNT_PATH[:ro|:rw], ' +
+	'e.g. hf://datasets/org/dataset:/data:ro or hf://buckets/org/bucket:/output.';
+const HF_VOLUME_TYPES: Record<string, JobVolumeType> = {
+	models: 'model',
+	datasets: 'dataset',
+	spaces: 'space',
+	buckets: 'bucket',
+};
+const SINGULAR_VOLUME_TYPES = new Set(['model', 'dataset', 'space', 'bucket']);
 
 function isEnvToken(entry: unknown): entry is EnvToken {
 	return Boolean(entry && typeof entry === 'object' && (entry as EnvToken).type === 'env');
@@ -50,9 +61,7 @@ export function parseTimeout(timeout: string): number {
 		if (!isNaN(seconds)) {
 			return seconds;
 		}
-		throw new Error(
-			`Invalid timeout format: "${timeout}". Use format like "5m", "2h", "30s", or plain seconds.`
-		);
+		throw new Error(`Invalid timeout format: "${timeout}". Use format like "5m", "2h", "30s", or plain seconds.`);
 	}
 
 	const value = parseFloat(match[1]);
@@ -93,7 +102,7 @@ export function parseCommand(command: string | string[]): { command: string[]; a
 	}
 
 	// Parse the command string using shell-quote for POSIX-compliant parsing
-	const parsed = parseShellArgs<EnvToken>(command, key => ({ type: 'env', key }));
+	const parsed = parseShellArgs<EnvToken>(command, (key) => ({ type: 'env', key }));
 
 	// Convert parsed result to string array
 	// shell-quote can return various types (strings, objects for operators, etc.)
@@ -121,6 +130,82 @@ export function parseCommand(command: string | string[]): { command: string[]; a
 	return { command: stringArgs, arguments: [] };
 }
 
+function invalidVolume(rawSpec: string, message: string): Error {
+	return new Error(`Invalid volume "${rawSpec}". ${message} ${VOLUME_FORMAT_HELP}`);
+}
+
+function parseVolume(rawSpec: string): JobVolume {
+	let spec = rawSpec;
+	let readOnly: boolean | undefined;
+
+	if (spec.endsWith(':ro')) {
+		readOnly = true;
+		spec = spec.slice(0, -3);
+	} else if (spec.endsWith(':rw')) {
+		readOnly = false;
+		spec = spec.slice(0, -3);
+	}
+
+	if (!spec.startsWith(HF_VOLUME_PREFIX)) {
+		throw invalidVolume(rawSpec, `Volume source must start with "${HF_VOLUME_PREFIX}".`);
+	}
+
+	const body = spec.slice(HF_VOLUME_PREFIX.length);
+	const mountSeparator = body.lastIndexOf(':/');
+	if (mountSeparator === -1) {
+		throw invalidVolume(rawSpec, 'Missing mount path.');
+	}
+
+	const sourcePart = body.slice(0, mountSeparator);
+	const mountPath = body.slice(mountSeparator + 1);
+	if (!sourcePart) {
+		throw invalidVolume(rawSpec, 'Missing Hub source before mount path.');
+	}
+	if (!mountPath.startsWith('/') || mountPath === '/') {
+		throw invalidVolume(rawSpec, `Mount path must be a non-empty absolute path, got "${mountPath}".`);
+	}
+
+	const segments = sourcePart.split('/');
+	const firstSegment = segments[0];
+	if (!firstSegment) {
+		throw invalidVolume(rawSpec, 'Missing Hub source type or owner.');
+	}
+	if (SINGULAR_VOLUME_TYPES.has(firstSegment)) {
+		throw invalidVolume(rawSpec, `Type prefix must be plural, got "${firstSegment}/".`);
+	}
+
+	const explicitType = HF_VOLUME_TYPES[firstSegment];
+	const type = explicitType ?? 'model';
+	const locationSegments = explicitType ? segments.slice(1) : segments;
+	if (locationSegments.length < 2 || !locationSegments[0] || !locationSegments[1]) {
+		throw invalidVolume(rawSpec, 'Hub source must include OWNER/NAME.');
+	}
+
+	const source = `${locationSegments[0]}/${locationSegments[1]}`;
+	const path = locationSegments.slice(2).join('/') || undefined;
+	const volume: JobVolume = { type, source, mountPath };
+
+	if (readOnly !== undefined) {
+		volume.readOnly = readOnly;
+	}
+	if (path) {
+		volume.path = path;
+	}
+
+	return volume;
+}
+
+/**
+ * Parse hf:// volume mount strings into the Jobs API payload shape.
+ */
+export function parseVolumes(volumes?: string[]): JobVolume[] | undefined {
+	if (!volumes || volumes.length === 0) {
+		return undefined;
+	}
+
+	return volumes.map(parseVolume);
+}
+
 /**
  * Replace HF token placeholder with actual token if available
  */
@@ -162,6 +247,7 @@ export function createJobSpec(args: {
 	secrets?: Record<string, string>;
 	timeout?: string;
 	hfToken?: string;
+	volumes?: string[];
 }): JobSpec {
 	// Validate required fields
 	if (!args.image) {
@@ -176,6 +262,7 @@ export function createJobSpec(args: {
 	const timeoutSeconds = args.timeout ? parseTimeout(args.timeout) : undefined;
 	const environment = transformEnvMap(args.env, args.hfToken) || {};
 	const secrets = transformEnvMap(args.secrets, args.hfToken) || {};
+	const volumes = parseVolumes(args.volumes);
 
 	const spec: JobSpec = {
 		...imageSource,
@@ -186,6 +273,9 @@ export function createJobSpec(args: {
 		secrets,
 		timeoutSeconds,
 	};
+	if (volumes) {
+		spec.volumes = volumes;
+	}
 
 	return spec;
 }

package/src/jobs/jobs-tool.ts

@@ -337,6 +337,42 @@ Manage compute jobs on Hugging Face infrastructure.
 
 ${renderExampleSection('Run a simple job', 'run')}${renderExampleSection('Run a Python script with UV', 'uv')}
 
+## Deep Hub Dataset/Repo Analysis
+
+Use Jobs for deep analysis prompts involving Hugging Face datasets, models, Spaces, repos, traces, or large Hub files—especially when the user asks to "analyze", "find trends", "process all rows/files", "run a complete analysis", "take your time", or "install/use Python libraries".
+
+Recommended workflow:
+1. Inspect the repo with \`hub_repo_details\` for schema, splits, and parquet URLs.
+2. Run \`operation: "uv"\` with a self-contained Python script; do not call \`{"operation": "uv"}\` by itself except to request help.
+3. Always put third-party packages in \`with_deps\`; do not assume packages like \`pandas\`, \`polars\`, \`pyarrow\`, \`datasets\`, or \`huggingface_hub\` are installed. Prefer \`with_deps\` over relying on inline PEP 723 script metadata.
+4. Prefer converted parquet URLs for Hub datasets when available; they are often more reliable for mixed JSONL/session repos than \`datasets.load_dataset(...)\`.
+5. Print the final report at the end of the job. If the initial response only shows installation logs or partial output, call \`logs\` with the exact returned job ID and a larger \`tail\`, e.g. \`{"tail": 500}\`.
+6. Jobs do not automatically inherit the MCP server's Hugging Face token inside the container. For private/gated data or uploads, pass \`secrets: { "HF_TOKEN": "$HF_TOKEN" }\`.
+
+Example:
+\`\`\`json
+{
+  "operation": "uv",
+  "args": {
+    "with_deps": ["polars", "pyarrow", "huggingface_hub"],
+    "timeout": "60m",
+    "flavor": "cpu-upgrade",
+    "script": "import polars as pl\\nurl = 'PARQUET_URL_FROM_HUB_REPO_DETAILS'\\ndf = pl.read_parquet(url)\\nprint(df.shape)\\nprint(df.head())"
+  }
+}
+\`\`\`
+
+If output is incomplete, fetch more logs:
+\`\`\`json
+{
+  "operation": "logs",
+  "args": {
+    "job_id": "JOB_ID_FROM_RUN_RESPONSE",
+    "tail": 500
+  }
+}
+\`\`\`
+
 ## Hardware Flavors
 
 ${HARDWARE_FLAVORS_SECTION}
@@ -357,6 +393,28 @@ ${HARDWARE_FLAVORS_SECTION}
 - Include newline characters directly in the argument (e.g., \`"first line\\nsecond line"\`)
 - UV inline scripts are automatically base64-decoded inside the container; just send the raw script text
 
+## Volumes
+
+Attach Hub repositories or buckets into the job container with \`hf://\` volume URLs.
+
+Format: \`hf://[TYPE/]OWNER/NAME[/PATH]:/MOUNT_PATH[:ro|:rw]\`
+
+- \`TYPE\` is one of \`models\`, \`datasets\`, \`spaces\`, or \`buckets\`; omitted type defaults to models.
+- \`OWNER/NAME\` source IDs are required.
+- \`:ro\` and \`:rw\` are optional; backend defaults are preserved when omitted.
+
+Example:
+\`\`\`json
+{
+  "operation": "run",
+  "args": {
+    "image": "python:3.12",
+    "command": ["python", "-c", "import os; print(os.listdir('/data'))"],
+    "volumes": ["hf://datasets/org/dataset:/data:ro"]
+  }
+}
+\`\`\`
+
 ### Show command-specific help
 Call this tool with:
 \`\`\`json
@@ -378,9 +436,8 @@ Call this tool with:
 export const HF_JOBS_TOOL_CONFIG = {
 	name: 'hf_jobs',
 	description:
-		'Manage Hugging Face CPU/GPU compute jobs. Run commands in Docker containers, ' +
-		'execute Python scripts with UV. List, schedule and monitor jobs/logs. ' +
-		'Call this tool with no operation for full usage instructions and examples. ',
+		'Remote compute for Hugging Face workflows. Run Python/UV or Docker jobs to deeply analyze Hub datasets, repos, traces, models, and large files; compute trends/statistics; run batch inference/evaluation; or perform long-running work with installed libraries. ' +
+		'Use for dataset/repo analysis prompts when local chat inspection is insufficient. Includes submit, logs, inspect, cancel, schedule, and volume mounting.',
 	schema: z.object({
 		operation: z
 			.enum(OPERATION_NAMES)

package/src/jobs/types.ts

@@ -60,6 +60,20 @@ export interface JobOwner {
 	type: 'user' | 'org';
 }
 
+/**
+ * Hugging Face Hub volume mounted into a Job container.
+ */
+export type JobVolumeType = 'bucket' | 'model' | 'dataset' | 'space';
+
+export interface JobVolume {
+	type: JobVolumeType;
+	source: string;
+	mountPath: string;
+	revision?: string;
+	readOnly?: boolean;
+	path?: string;
+}
+
 /**
  * Job information from API
  * Based on OpenAPI schema
@@ -97,6 +111,7 @@ export interface JobSpec {
 	secrets?: Record<string, string>;
 	flavor: string;
 	timeoutSeconds?: number;
+	volumes?: JobVolume[];
 }
 
 /**
@@ -164,6 +179,14 @@ export const runArgsSchema = commonArgsSchema.extend({
 		.optional()
 		.describe('Secrets as key-value pairs. Use HF_TOKEN=$HF_TOKEN to include your token'),
 	timeout: z.string().optional().describe('Max duration (e.g., "5m", "2h", "30s"). Default: 30m').default('30m'),
+	volumes: z
+		.array(z.string())
+		.optional()
+		.describe(
+			'Volume mounts using hf:// URLs. Format: hf://TYPE/OWNER/NAME[/PATH]:/MOUNT_PATH[:ro|:rw]. ' +
+				'TYPE is models, datasets, spaces, or buckets. ' +
+				'Examples: ["hf://datasets/org/ds:/data:ro", "hf://buckets/org/b:/output"].'
+		),
 	detach: z
 		.boolean()
 		.optional()
@@ -187,7 +210,18 @@ export const uvArgsSchema = commonArgsSchema.extend({
 		.optional()
 		.describe('Secrets as key-value pairs. Use HF_TOKEN=$HF_TOKEN to include your token'),
 	timeout: z.string().optional().default('30m').describe('Max duration'),
-	detach: z.boolean().optional().default(false).describe('If true, return immediately with job ID. If false (default), tail logs for up to 10 seconds.'),
+	volumes: z
+		.array(z.string())
+		.optional()
+		.describe(
+			'Volume mounts using hf:// URLs. Format: hf://TYPE/OWNER/NAME[/PATH]:/MOUNT_PATH[:ro|:rw]. ' +
+				'TYPE is models, datasets, spaces, or buckets.'
+		),
+	detach: z
+		.boolean()
+		.optional()
+		.default(false)
+		.describe('If true, return immediately with job ID. If false (default), tail logs for up to 10 seconds.'),
 });
 
 // PS command args

package/src/model-detail.ts

@@ -259,7 +259,7 @@ export class ModelDetailTool {
 
 			// Fetch and append README content if requested
 			if (includeReadme) {
-				const readmeContent = await fetchReadmeContent(modelDetails.name, 'models', false);
+				const readmeContent = await fetchReadmeContent(modelDetails.name, 'models');
 				if (readmeContent) {
 					const result = formatModelDetails(modelDetails);
 					result.formatted += '\n\n## README\n<modelcard-readme>\n\n' + readmeContent.trim() + '\n</modelcard-readme>';

package/src/readme-utils.ts

@@ -5,21 +5,16 @@
 import { fetchWithProfile, NETWORK_FETCH_PROFILES } from './network/fetch-profile.js';
 
 // Maximum number of characters to include from a README
-const DEFAULT_MAX_README_CHARS = 10_000;
+const DEFAULT_MAX_README_CHARS = 40_000;
 
 /**
  * Fetches README content from a Hugging Face repository
  *
  * @param repoName The resolved repository name (e.g., 'rajpurkar/squad', 'openai-community/gpt2')
  * @param type The repository type ('models' or 'datasets')
- * @param includeYaml Whether to include YAML frontmatter (default: false)
  * @returns Promise<string | null> The README content or null if not found/error
  */
-export async function fetchReadmeContent(
-	repoName: string,
-	type: 'models' | 'datasets',
-	includeYaml: boolean = false
-): Promise<string | null> {
+export async function fetchReadmeContent(repoName: string, type: 'models' | 'datasets'): Promise<string | null> {
 	try {
 		// Construct the URL based on repository type
 		const baseUrl =
@@ -39,11 +34,6 @@ export async function fetchReadmeContent(
 
 		let content = await response.text();
 
-		// If includeYaml is false, strip YAML frontmatter
-		if (!includeYaml) {
-			content = stripYamlFrontmatter(content);
-		}
-
 		// Truncate overly long READMEs to a sensible default size
 		if (content.length > DEFAULT_MAX_README_CHARS) {
 			const truncated = content.slice(0, DEFAULT_MAX_README_CHARS);
@@ -62,23 +52,3 @@ export async function fetchReadmeContent(
 		return null;
 	}
 }
-
-/**
- * Strips YAML frontmatter from markdown content
- *
- * @param content The full markdown content
- * @returns The content with YAML frontmatter removed
- */
-function stripYamlFrontmatter(content: string): string {
-	// Match YAML frontmatter: starts with ---, ends with ---
-	const yamlPattern = /^(\s*---[\r\n]+)([\S\s]*?)([\r\n]+---(\r\n|\n|$))/;
-	const match = content.match(yamlPattern);
-
-	if (match) {
-		// Return everything after the closing ---
-		return content.substring(match[0].length);
-	}
-
-	// No YAML frontmatter found, return original content
-	return content;
-}