jina-ai

name	jina-ai
description	Web search in specific languages/regions (local community content), academic papers (arXiv/SSRN), PDF extraction, BibTeX, and image search via Jina AI. This skill should be used when searching non-English content, finding academic papers, extracting figures from PDFs, searching for images, or user says "search in Chinese", "find papers on arXiv", "search for images of", "get BibTeX". Prefer this over WebSearch when content is non-English, academic, or media (PDF/image).
allowed-tools	["Bash(*mcpcall.py*:*)"]

Jina AI

Call Jina MCP tools via scripts/mcpcall.py for web content extraction, search, academic research, embeddings-based NLP, and visual capture.

Setup

Requires JINA_API_KEY environment variable (get one at jina.ai/api-key):

export JINA_API_KEY=<key>

Web Reading ⚠️ Unreliable — use defuddle or WebFetch instead

read_url ❌

Extract web page content as clean markdown. Supports single URL or array of URLs.

• url (required): URL string or array of URLs
• withAllLinks: extract all hyperlinks as structured data
• withAllImages: extract all images as structured data

scripts/mcpcall.py read_url url:"https://example.com"
scripts/mcpcall.py read_url url:"https://example.com" withAllLinks:true

parallel_read_url ❌

Read up to 5 URLs in parallel for batch extraction.

• urls (required): array of {url, withAllLinks?, withAllImages?} objects
• timeout: milliseconds (default 30000)

scripts/mcpcall.py parallel_read_url --args '{"urls": [{"url": "https://a.com"}, {"url": "https://b.com"}]}'

Web Search

search_web

Search the web for current information. Supports single query or array of queries for parallel search.

• query (required): search terms (string or array)
• num: max results 1-100 (default 30)
• tbs: time filter — qdr:h (hour), qdr:d (day), qdr:w (week), qdr:m (month), qdr:y (year)
• gl: country code (e.g. cn)
• hl: language code (e.g. zh-cn)
• location: location string (e.g. Shanghai)

scripts/mcpcall.py search_web query:"search terms" num:10
scripts/mcpcall.py search_web query:"关键词" gl:cn hl:zh-cn
scripts/mcpcall.py search_web query:"recent news" tbs:qdr:w

parallel_search_web

Run up to 5 web searches in parallel for broader coverage.

• searches (required): array of {query, num?, tbs?, gl?, hl?, location?}
• timeout: milliseconds (default 30000)

scripts/mcpcall.py parallel_search_web --args '{"searches": [{"query": "topic A"}, {"query": "topic B", "num": 5}]}'

search_images

Search for images across the web (like Google Images). Returns base64 JPEG by default.

• query (required): image search terms
• return_url: set true to get URLs instead of base64
• tbs, gl, hl, location: same as search_web

scripts/mcpcall.py search_images query:"neural network diagram"
scripts/mcpcall.py search_images query:"logo" return_url:true

Academic Research

search_arxiv

Search arXiv for academic papers in STEM fields.

• query (required): search terms, author names, or topics (string or array)
• num: max results 1-100 (default 30)
• tbs: time filter

scripts/mcpcall.py search_arxiv query:"transformer attention" num:10
scripts/mcpcall.py search_arxiv query:"reinforcement learning" tbs:qdr:m

parallel_search_arxiv

Run up to 5 arXiv searches in parallel for comprehensive coverage.

• searches (required): array of {query, num?, tbs?}
• timeout: milliseconds (default 30000)

scripts/mcpcall.py parallel_search_arxiv --args '{"searches": [{"query": "topic A"}, {"query": "topic B"}]}'

search_ssrn

Search SSRN for social science, economics, law, finance papers.

• query (required): search terms (string or array)
• num: max results 1-100 (default 30)
• tbs: time filter

scripts/mcpcall.py search_ssrn query:"market microstructure" num:10

parallel_search_ssrn

Run up to 5 SSRN searches in parallel.

• searches (required): array of {query, num?, tbs?}
• timeout: milliseconds (default 30000)

scripts/mcpcall.py parallel_search_ssrn --args '{"searches": [{"query": "topic A"}, {"query": "topic B"}]}'

search_bibtex

Search DBLP + Semantic Scholar, return BibTeX citations.

• query (required): paper title, topic, or keywords
• author: filter by author name
• year: minimum publication year
• num: max results 1-50 (default 10)

scripts/mcpcall.py search_bibtex query:"attention is all you need"
scripts/mcpcall.py search_bibtex query:"deep learning" author:Hinton year:2020 num:5

PDF & Screenshots

extract_pdf

Extract figures, tables, and equations from PDFs using layout detection.

• id: arXiv paper ID (e.g. 2301.12345)
• url: direct PDF URL
• type: filter by figure, table, equation (comma-separated)
• max_edge: max image edge size in px (default 1024)

scripts/mcpcall.py extract_pdf id:2301.12345
scripts/mcpcall.py extract_pdf url:"https://example.com/paper.pdf" type:figure

capture_screenshot_url ❌ Use agent-browser instead

Capture web page screenshots as base64 JPEG.

• url (required): page URL
• firstScreenOnly: true for viewport only (faster), false for full page
• return_url: true to get URL instead of base64

scripts/mcpcall.py capture_screenshot_url url:"https://example.com"
scripts/mcpcall.py capture_screenshot_url url:"https://example.com" firstScreenOnly:true

NLP & Embeddings

classify_text

Classify texts into user-defined labels using Jina embeddings.

• texts (required): array of strings to classify
• labels (required): array of label strings
• model: embedding model (default jina-embeddings-v5-text-small)

scripts/mcpcall.py classify_text --args '{"texts": ["great product", "terrible"], "labels": ["positive", "negative", "neutral"]}'

sort_by_relevance

Rerank documents by relevance to a query using Jina Reranker.

• query (required): the query to rank against
• documents (required): array of document texts
• top_n: max results to return

scripts/mcpcall.py sort_by_relevance --args '{"query": "machine learning", "documents": ["doc1 text", "doc2 text"], "top_n": 5}'

deduplicate_strings

Select top-k semantically unique strings from a list.

• strings (required): array of strings
• k: number to return (auto-optimized if omitted)

scripts/mcpcall.py deduplicate_strings --args '{"strings": ["hello world", "hi world", "goodbye"]}'

deduplicate_images

Select top-k visually unique images using CLIP v2 embeddings.

• images (required): array of image URLs or base64 strings
• k: number to return (auto-optimized if omitted)

scripts/mcpcall.py deduplicate_images --args '{"images": ["https://a.com/1.jpg", "https://a.com/2.jpg"]}'

expand_query

Rewrite a search query into multiple expanded variants for deeper research.

• query (required): the query to expand

scripts/mcpcall.py expand_query query:"machine learning optimization"

Utility

primer

Get current session context (time, location, network) for localized responses. No parameters.

scripts/mcpcall.py primer

guess_datetime_url

Guess when a web page was last updated/published.

• url (required): page URL

scripts/mcpcall.py guess_datetime_url url:"https://example.com/article"

search_jina_blog

Search Jina AI's official blog and news.

• query (required): search terms (string or array)
• num: max results 1-100 (default 30)
• tbs: time filter

scripts/mcpcall.py search_jina_blog query:"embeddings" num:10

show_api_key

Show the current Jina API key for this session. No parameters.

scripts/mcpcall.py show_api_key

Tool Selection Guide

Scenario	Tool	Notes
Read a web page	defuddle or WebFetch
General web search	WebSearch (built-in)
Region/language-specific search	search_web with gl/hl	e.g. gl:jp hl:ja for Japanese results
Find STEM papers	search_arxiv	Jina's strength
Find social science / finance papers	search_ssrn	Jina's strength
Get BibTeX citations	search_bibtex	Jina's strength
Extract figures from a paper	extract_pdf	Jina's strength
Find images	search_images
Categorize text into labels	classify_text
Rank documents by relevance	sort_by_relevance
Remove duplicate content	deduplicate_strings / deduplicate_images

Tips

• Use expand_query before parallel_search_web or parallel_search_arxiv to generate diverse queries for thorough research.
• Parallel variants accept up to 5 items — use them for batch work.
• Set tbs:qdr:w to restrict results to the past week for time-sensitive queries.
• For Chinese-language results, set gl:cn hl:zh-cn on search tools.
• scripts/mcpcall.py --list jina to see all available tools.