visual-hunt-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CHROME_CDP_ENDPOINT | No | Chrome remote-debugging endpoint | http://127.0.0.1:9222 |
| VISUAL_HUNT_TIMEOUT_MS | No | Per-page navigation/extraction timeout | 20000 |
| VISUAL_HUNT_OUTPUT_ROOT | No | Root directory for downloaded images and manifests | process.cwd() |
| VISUAL_HUNT_AUTO_REGISTER | No | Set to 1 to rewrite ~/.claude.json with this server entry on every startup | |
| VISUAL_HUNT_MANUAL_PAGE_MAX | No | Hard cap on manual-page registry size; oldest evicted first | 50 |
| VISUAL_HUNT_MANUAL_PAGE_TTL_MS | No | TTL for entries in the manual-page registry; getManualPage refreshes | 1800000 |
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": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_google_imagesA | Search Google Images (udm=2) for visually strong poster and wallpaper candidates using the connected local Chrome session. Default quality is 'normal' — pass 'cinematic' or 'poster' only when you actually want a stylistic suffix appended to the query (verbose suffixes can push Google to zero-results pages). |
| search_wallpaper_sitesA | Search supported wallpaper, photography, archive, and open-license image sources using direct site adapters with Google fallback. Pass |
| open_instagram_visual_searchA | Open Instagram in the connected local Chrome session for manual-assisted visual review and extract only visible images. When pageState.status is |
| open_xiaohongshu_visual_searchA | Open Xiaohongshu search results in the connected local Chrome session and extract visible images. Best path for Chinese-language queries. When pageState reports a login wall, candidates is empty — log in via Chrome and call extract_manual_page_candidates. |
| open_pinterest_visual_searchA | Open Pinterest pin search results in the connected local Chrome session and extract visible image candidates. Pinterest occasionally serves a login wall — log in via the connected Chrome window and call extract_manual_page_candidates to re-extract. |
| open_google_maps_visual_searchA | Open Google Maps in the connected local Chrome session for safe, manual-assisted visual inspiration review. Map tile URLs (maps/vt/...) are filtered out automatically; remaining Google-hosted preview URLs cannot be downloaded by this MCP. |
| open_url_and_extract_imagesA | Open any http or https URL in the connected local Chrome session and extract visible image candidates. javascript:, file:, and other schemes are rejected. |
| extract_manual_page_candidatesA | Re-extract visible image candidates from a previously opened manual-assisted page after you log in, scroll, or dismiss overlays. The manualPageId comes from a prior open_*_visual_search call. Each access refreshes the TTL. |
| close_manual_pageA | Close a previously opened manual-assisted page and remove it from the in-memory registry. Returns {closed:false} if the manualPageId was unknown (already closed or expired); not an error. |
| save_candidates_jsonA | Save candidate metadata under the current MCP working directory without downloading image files. |
| download_candidate_imagesA | Download candidate images into a project folder under the current MCP working directory, using the connected browser session for cookies and referer when needed. Google-hosted previews and Google Maps imagery are blocked. |
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 11 tools
Each tool targets a distinct action or platform: opening specific sites (Google Maps, Instagram, Pinterest, Xiaohongshu, generic URL), extracting or re-extracting candidates, downloading or saving images, and searching via Google or wallpaper sites. No two tools have overlapping purposes, and descriptions clarify their unique roles.
All tool names follow a consistent verb_noun pattern in snake_case (e.g., 'close_manual_page', 'download_candidate_images', 'open_google_maps_visual_search'). There are no deviations or mixed styles, ensuring predictability for agents.
With 11 tools, the server is well-scoped for its purpose of visual hunting. Each tool serves a clear function without redundancy, and the count is neither too thin nor too heavy for the domain.
The tool set covers the full workflow: opening visual sources (platform-specific and generic), extracting and re-extracting candidates, downloading images, and saving metadata. There are no obvious missing operations like a 'list downloads' tool, but the core lifecycle is complete and agents can accomplish tasks without dead ends.