Jina AI - Web Reader, Search and Deep Search
Verified@adhishthite
npx machina-cli add skill @adhishthite/jina-ai --openclawJina AI — Reader, Search & DeepSearch
Web reading and search powered by Jina AI. Requires JINA_API_KEY environment variable.
Trust & Privacy: By using this skill, URLs and queries are transmitted to Jina AI (jina.ai). Only install if you trust Jina with your data.
Model Invocation: This skill may be invoked autonomously by the model without explicit user trigger (standard for integration skills). If you prefer manual-only invocation, disable model invocation in your OpenClaw skill settings.
Get your API key: https://jina.ai/ → Dashboard → API Keys
External Endpoints
This skill makes HTTP requests to the following external endpoints only:
| Endpoint | URL Pattern | Purpose |
|---|---|---|
| Reader API | https://r.jina.ai/{url} | Sends URL content request to Jina for conversion to markdown |
| Search API | https://s.jina.ai/{query} | Sends search query to Jina for web search results |
| DeepSearch API | https://deepsearch.jina.ai/v1/chat/completions | Sends research question to Jina for multi-step research |
No other external network calls are made by this skill.
Security & Privacy
- Authentication: Only your
JINA_API_KEYis transmitted to Jina's servers (viaAuthorizationheader) - Data sent: URLs and search queries you provide are sent to Jina's servers for processing
- Local files: No local files are read or transmitted by this skill
- Local storage: No data is stored locally beyond stdout output
- Environment access: Scripts only access the
JINA_API_KEYenvironment variable; no other env vars are read - Cookies: Cookies are not forwarded by default; the
X-Set-Cookieheader is available for authenticated content but is opt-in only
Endpoints
| Endpoint | Base URL | Purpose |
|---|---|---|
| Reader | https://r.jina.ai/{url} | Convert any URL → clean markdown |
| Search | https://s.jina.ai/{query} | Web search with LLM-friendly results |
| DeepSearch | https://deepsearch.jina.ai/v1/chat/completions | Multi-step research agent |
All endpoints accept Authorization: Bearer $JINA_API_KEY.
Reader API (r.jina.ai)
Fetches any URL and returns clean, LLM-friendly content. Works with web pages, PDFs, and JS-heavy sites.
Basic Usage
# Plain text output
curl -s "https://r.jina.ai/https://example.com" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: text/plain"
# JSON output (includes url, title, content, timestamp)
curl -s "https://r.jina.ai/https://example.com" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: application/json"
Or use the helper script: scripts/jina-reader.sh <url> [--json]
Parameters (via headers or query params)
Content Control
| Header | Query Param | Values | Default | Description |
|---|---|---|---|---|
X-Respond-With | respondWith | content, markdown, html, text, screenshot, pageshot, vlm, readerlm-v2 | content | Output format |
X-Retain-Images | retainImages | none, all, alt, all_p, alt_p | all | Image handling |
X-Retain-Links | retainLinks | none, all, text, gpt-oss | all | Link handling |
X-With-Generated-Alt | withGeneratedAlt | true/false | false | Auto-caption images |
X-With-Links-Summary | withLinksSummary | true | - | Append links section |
X-With-Images-Summary | withImagesSummary | true/false | false | Append images section |
X-Token-Budget | tokenBudget | number | - | Max tokens for response |
CSS Selectors
| Header | Query Param | Description |
|---|---|---|
X-Target-Selector | targetSelector | Only extract matching elements |
X-Wait-For-Selector | waitForSelector | Wait for elements before extracting |
X-Remove-Selector | removeSelector | Remove elements before extraction |
Browser & Network
| Header | Query Param | Description |
|---|---|---|
X-Timeout | timeout | Page load timeout (1-180s) |
X-Respond-Timing | respondTiming | When page is "ready" (html, network-idle, etc.) |
X-No-Cache | noCache | Bypass cached content |
X-Proxy | proxy | Country code or auto for proxy |
X-Set-Cookie | setCookies | Forward cookies for authenticated content |
Common Patterns
# Extract main content, remove navigation elements
curl -s "https://r.jina.ai/https://example.com/article" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "X-Retain-Images: none" \
-H "X-Remove-Selector: nav, footer, .sidebar, .ads" \
-H "Accept: text/plain"
# Extract specific section
curl -s "https://r.jina.ai/https://example.com" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "X-Target-Selector: article.main-content"
# Parse a PDF
curl -s "https://r.jina.ai/https://example.com/paper.pdf" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: text/plain"
# Wait for dynamic content
curl -s "https://r.jina.ai/https://spa-app.com" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "X-Wait-For-Selector: .loaded-content" \
-H "X-Respond-Timing: network-idle"
Search API (s.jina.ai)
Web search returning LLM-friendly results with full page content.
Basic Usage
# Plain text
curl -s "https://s.jina.ai/your+search+query" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: text/plain"
# JSON
curl -s "https://s.jina.ai/your+search+query" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: application/json"
Or use the helper script: scripts/jina-search.sh "<query>" [--json]
Search Parameters
| Param | Values | Description |
|---|---|---|
site | domain | Limit to specific site |
type | web, images, news | Search type |
num / count | 0-20 | Number of results |
gl | country code | Geo-location (e.g. us, in) |
filetype | extension | Filter by file type |
intitle | string | Must appear in title |
All Reader parameters also work on search results.
Common Patterns
# Site-scoped search
curl -s "https://s.jina.ai/OpenAI+GPT-5?site=reddit.com" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: text/plain"
# News search
curl -s "https://s.jina.ai/latest+AI+news?type=news&num=5" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Accept: application/json"
# Search for PDFs
curl -s "https://s.jina.ai/machine+learning+survey?filetype=pdf&num=5" \
-H "Authorization: Bearer $JINA_API_KEY"
DeepSearch
Multi-step research agent that combines search + reading + reasoning. OpenAI-compatible chat completions API.
curl -s "https://deepsearch.jina.ai/v1/chat/completions" \
-H "Authorization: Bearer $JINA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "jina-deepsearch-v1",
"messages": [{"role": "user", "content": "Your research question here"}],
"stream": false
}'
Or use the helper script: scripts/jina-deepsearch.sh "<question>"
Use for complex research requiring multiple sources and reasoning chains.
Helper Scripts
| Script | Purpose |
|---|---|
scripts/jina-reader.sh | Read any URL as markdown |
scripts/jina-search.sh | Web search |
scripts/jina-deepsearch.sh | Deep multi-step research |
scripts/jina-reader.py | Python reader (no deps beyond stdlib) |
Rate Limits
- Free (no key): 20 RPM
- With API key: Higher limits, token-based pricing
API Docs
- Reader: https://jina.ai/reader
- Search: https://s.jina.ai/docs
- OpenAPI specs: https://r.jina.ai/openapi.json | https://s.jina.ai/openapi.json
When to Use
| Need | Use |
|---|---|
| Fetch a URL as markdown | Reader — better than web_fetch for JS-heavy sites |
| Web search | Search — LLM-friendly results |
| Complex multi-source research | DeepSearch |
| Parse a PDF from URL | Reader — pass PDF URL directly |
| Screenshot a page | Reader with X-Respond-With: screenshot |
| Extract structured data | Reader with jsonSchema param |
Overview
Jina AI — Reader, Search & DeepSearch enables fetching clean markdown from URLs, performing web searches, and conducting multi-step research with Jina AI. It uses three endpoints: Reader (r.jina.ai/{url}) for URL-to-markdown, Search (s.jina.ai/{query}) for web results, and DeepSearch (deepsearch.jina.ai/v1/chat/completions) for multi-step investigations. A JINA_API_KEY is required and queries are transmitted to Jina AI as described in the privacy notes.
How This Skill Works
Requests are authenticated with Authorization: Bearer $JINA_API_KEY and routed to three endpoints: Reader (r.jina.ai/{url}) to convert pages into clean markdown or text, Search (s.jina.ai/{query}) to return LLMed-friendly results, and DeepSearch for multi-step research. The Reader API handles web pages, PDFs, and JS-heavy sites, while DeepSearch enables structured, multi-turn inquiries. Output formats and content behavior are controllable via headers and query parameters described in the Reader API docs.
When to Use It
- Extract clean content from a web page for summarization or quick review
- Perform targeted web searches with results optimized for language models
- Conduct deep, multi-step research on a topic using a conversational workflow
- Power a live web-content fetch feature in an AI assistant or chatbot
- Fetch content from PDFs or JS-heavy sites that are hard to scrape with basic bots
Quick Start
- Step 1: Get your API key from the Jina dashboard (JINA_API_KEY)
- Step 2: Fetch a URL with Reader for plain text: curl -s "https://r.jina.ai/https://example.com" -H "Authorization: Bearer $JINA_API_KEY" -H "Accept: text/plain"
- Step 3: Optional JSON output or use the helper script: # JSON output curl -s "https://r.jina.ai/https://example.com" -H "Authorization: Bearer $JINA_API_KEY" -H "Accept: application/json" # Or use the helper script scripts/jina-reader.sh https://example.com --json
Best Practices
- Use Reader for URL-to-markdown/plain text extraction to prepare content for downstream tasks
- Choose output formats with Accept headers or X-Respond-With (e.g., markdown, content, json) to fit your workflow
- Always include the Authorization: Bearer $JINA_API_KEY header for secure requests
- Leverage DeepSearch for complex, multi-step research rather than simple queries
- Be mindful of privacy: URLs and queries are transmitted to Jina AI; review trust/privacy notes before use
Example Use Cases
- Summarize a long article by fetching it via Reader and extracting key points as markdown
- Run a topic search and collect top results in a structured, model-friendly format
- Perform a multi-source DeepSearch to triangulate information on a complex topic
- Convert a PDF page into readable markdown for quick review
- Integrate into an AI assistant to fetch live web content on user request