Skip to main content
Give your AI agents access to accurate, real-time web data. This guide covers how to integrate Linkup into any agent environment — from MCP-based setups to custom-built pipelines — and how to configure your agent to use Linkup effectively. You can give this page as context to your agent. We also publish a Linkup Skill — a reusable knowledge module that agents can install and reference automatically for better Linkup API usage.

Install the Linkup Skill

One command: npx skills add LinkupPlatform/skills — works with Claude Code, Cursor, Windsurf, and more.

​
1. System Prompt Snippet

Copy this into your agent’s system prompt, tool description, or CLAUDE.md file. It teaches any LLM how to use Linkup well. 
## Linkup Web Search — Usage Instructions

You have access to Linkup, a web search and content extraction API. Follow these rules:

### Before Writing a Query
Reason through three questions in order:
1. INPUTS: What do I already have? A URL? A company name? A topic?
   - If I have a URL → scrape it directly, don't search for it.
2. DATA LOCATION: Where does the data I need live?
   - In search snippets (facts, dates, names, short claims) → search is enough.
   - On full web pages (tables, specs, detailed content) → need to scrape.
   - Not sure → default to standard.
3. SEQUENCING: Do I need to chain steps?
   - Parallel searches only → standard.
   - Find URL then scrape it, or scrape then search again → deep.
   - Scrape multiple URLs → deep.

### Search Depth
- depth="standard" (€0.005): multiple parallel searches + scrape one provided URL. Cannot scrape multiple URLs. Cannot chain search → scrape.
- depth="deep" (€0.05): up to 10 iterative steps. Can scrape multiple URLs. Can chain search → scrape. Can run new searches based on extracted information.
- When uncertain, default to standard.

### Query Style
- Simple factual lookups → short keyword queries.
- Complex extraction → natural language instructions: what to find, where to look, what to extract.
- Broad research → "Run several searches with adjacent keywords." Works in standard.
- Don't ask Linkup to analyze or reason. Ask it to retrieve. You do the thinking.

### Scraping Rules
- Standard: scrape one URL provided in the prompt. That's it.
- Deep: scrape multiple URLs, scrape URLs discovered during search.

### Output Type
- outputType="searchResults": raw sources for you to process (default for agents).
- outputType="sourcedAnswer": natural language answer with citations.
- outputType="structured" + JSON schema: machine-parseable data.

### Fetch
- Use /fetch instead of /search for a single known URL.
- Always set renderJs=true unless the page is static HTML.
Where to put this: System prompt, CLAUDE.md, .cursorrules, MCP tool descriptions, or framework tool docstrings. It works everywhere.

​
2. Query Construction — Examples

Your query should tell Linkup what to retrieve, not what to think. The system prompt snippet above teaches the 3-step reasoning (inputs → data location → sequencing). Here’s how it plays out: 
Task: Get a company's pricing
Reasoning: no URL → need to find it → then scrape → sequential → deep
→ query: "Find the pricing page for {company}. Scrape it. Extract plan names, prices, and features."

Task: Get a company's latest funding round
Reasoning: no URL → answer lives in snippets → no chaining → standard
→ query: "Find {company}'s latest funding round amount and date"

Task: Extract data from a known URL
Reasoning: have URL → just scrape → no chaining → standard or /fetch
→ query: "Scrape https://example.com/pricing. Extract plan names, prices, and included features."

Task: Build an ICP from a company's web presence
Reasoning: no URL → need full pages → find then scrape multiple → deep
→ query: "Find and scrape {company}'s homepage, use case pages, and 2-3 recent blog posts. Extract: industries mentioned, company sizes referenced, job titles targeted, and pain points addressed."
For more query examples and patterns, see the Prompting Guide.

​
3. Output Types

For most agent use cases, use searchResults.
Output TypeReturnsBest For
searchResultsArray of {name, url, content}Agent-side reasoning, RAG, multi-source synthesis
sourcedAnswerAnswer + source citationsUser-facing chatbots, Q&A
structuredJSON matching your schemaCRM updates, data pipelines, enrichment
For details on structured output schemas, see API Reference.

​
4. /fetch Endpoint

Use /fetch instead of /search when your agent already has the exact URL. It’s faster, cheaper, and returns clean markdown. Always set renderJs: true (most sites need it). Full fetch documentation →

​
5. Integration by Environment

​
To Go Further

Need help? Reach out at [email protected], on Discord, or book a 15-minute call.