Social Image MCP Crawler
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| PYTHONUTF8 | No | Set to 1 to force UTF-8 encoding (recommended on Windows). | |
| OBJECT_MODEL | No | Local detection/segmentation model for object classification. | |
| VISION_MODEL | No | Local CLIP model for visual reranking, default openai/clip-vit-base-patch32. | openai/clip-vit-base-patch32 |
| DOUYIN_COOKIE | No | Douyin cookie for authentication. | |
| SEMANTIC_MODEL | No | Model for local semantic reranking, e.g., BAAI/bge-small-zh-v1.5. | |
| X_BEARER_TOKEN | No | X API v2 Bearer Token for X platform access. | |
| OBJECT_LABEL_MAP | No | JSON mapping custom labels to standard categories. | |
| PYTHONIOENCODING | No | Set to utf-8 for proper I/O encoding (recommended on Windows). | |
| GALLERY_DL_BINARY | No | Path to gallery-dl binary for X/Instagram/Weibo. | |
| INSTAGRAM_USER_ID | No | Instagram user ID for hashtag search. | |
| VISION_LOCAL_ONLY | No | Only use local vision models, default true. | true |
| MEDIA_CRAWLER_COMMAND | No | Command template for MediaCrawler bridge (XHS/Weibo). | |
| DOUYIN_BROWSER_POLL_MS | No | Polling interval for Douyin browser, default 400. | 400 |
| INSTAGRAM_ACCESS_TOKEN | No | Instagram Graph API access token. | |
| SEARCH_TIMEOUT_SECONDS | No | Global search timeout, default 55. | 55 |
| XHS_DOWNLOADER_COMMAND | No | Command template for XHS-Downloader bridge (XHS detail). | |
| PLATFORM_TIMEOUT_SECONDS | No | Per-platform timeout, default 50. | 50 |
| SEMANTIC_TIMEOUT_SECONDS | No | Timeout for semantic model, default 4. | 4 |
| DOUYIN_MEDIA_CRAWLER_FALLBACK | No | Enable MediaCrawler fallback for Douyin, default false. | false |
| OBJECT_FILTER_TIMEOUT_SECONDS | No | Timeout for object filtering, default 4. | 4 |
| SOURCE_FAILURE_COOLDOWN_SECONDS | No | Cooldown period after source failure, default 120. | 120 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_imagesC | Search social platforms by a keyword, content ID, or post URL. Creator queries such as douyin-user: or douyin-name: route to exact account retrieval; media_type=videos can return original video files for creator queries. |
| fetch_creator_imagesA | Download creator media from Douyin, Weibo or Bilibili, NOT a post ID. Douyin accepts an exact account handle/UID/sec_uid via creator_id, an exact nickname via creator_name, or a full profile URL via profile_url. If a Douyin handle or nickname cannot be found, retry with the complete https://www.douyin.com/user/ profile URL because it is more stable. Weibo and Bilibili accept numeric UID or full profile URL. media_type=images downloads image galleries, media_type=videos downloads original video files, and media_type=all returns both. By default this is fast account-only retrieval with no semantic or visual filtering. To keep only a content theme, provide content_query; the service classifies coarse objects such as person, clothing, landscape, scene, architecture and body regions, then applies include/exclude/required rules. Use filter_mode=optional to fall back when the local model is unavailable, or required to fail closed. Downloads are bounded by max_posts and max_images. Repeat with resume=true and the same options to continue pending media. |
| download_imagesB | Download ranked image candidates returned by search_images. Performs retries, image validation, minimum-size filtering and content-hash deduplication. |
| inspect_itemB | Inspect a single platform item by ID and return its image candidates. |
| list_platformsA | List supported platforms and their effective availability through recommended source projects or optional legacy adapters. |
| list_sourcesA | List recommended source projects and their configuration status. Sources are used for candidate recall; AI ranking is handled by this MCP server. |
| submit_feedbackB | Record whether a returned image matched the user's intent. This local feedback is used to rerank the same type of future results. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 7 tools
search_images and fetch_creator_images overlap heavily: search_images already routes creator queries (douyin-user:, douyin-name:) and fetch_creator_images also retrieves creator media, so the boundary is blurred. download_images vs fetch_creator_images both download media, and inspect_item vs search_images both return image candidates, though descriptions provide partial differentiation.
All tools follow a consistent verb_noun snake_case pattern (inspect_item, search_images, fetch_creator_images, download_images, list_platforms, list_sources, submit_feedback). No mixing of conventions or vague bare verbs.
Seven tools is well-scoped for a social image crawling server: search/inspect, creator fetch, download, two listing utilities, and feedback. Each tool appears to earn its place without redundancy bloat.
The surface covers discovery (search/inspect), retrieval (creator fetch), downloading, platform/source enumeration, and feedback. Minor gaps: no explicit way to configure or add sources (only list_sources), and no tool to manage adapters despite legacy adapter references.