# Agent Skill

A skill is a short instruction file your AI agent reads so it knows when and how to search news - the right tool, the right time window, the right filters - without being told each time.

> [!NOTE]
> Set up the [MCP Server](news-search-api-mcp) first. The MCP connection provides the tool; the skill teaches the agent to use it well.

## Install

Copy the skill file below to:

| Client | Path |
| --- | --- |
| Cursor | `~/.cursor/skills/webz-news-search/SKILL.md` |
| Claude Code | `~/.claude/skills/webz-news-search/SKILL.md` |

Restart your client after saving.

## The skill file

```markdown
---
name: webz-news-search
description: Search global news with the Webz.io news_search_by_webz MCP tool. Use for any question about news, current events, or recent developments around a topic, company, person, or place.
---

# Webz.io News Search

Use the `news_search_by_webz` MCP tool to answer news and current-events questions.

## When to use

- The user asks about news, recent events, announcements, or coverage of a topic.
- Fresh, sourced information matters more than general knowledge.

## How to call

- `query`: a natural-language topic or question ("EU AI regulation progress"), not keyword syntax.
- The default window is the last 7 days. For older news pass `days` (up to 30) or `allow_all_dates: true`.
- `k`: articles to return - default 10, max 50. Raise it for broad research, lower it for a quick answer.

## Filters (optional, values are case-insensitive)

- `language`: full names - english, chinese, hebrew, french, spanish, german, arabic, russian, japanese, korean.
- `country`: ISO-2 uppercase codes - "US", "GB", "DE", "IL".
- `sentiment`: positive, negative, neutral.
- `category`: one of 17 values - Arts, Culture and Entertainment; Crime, Law and Justice; Disaster and Accident; Economy, Business and Finance; Education; Environment; Health; Human Interest; Labor; Lifestyle and Leisure; Politics; Religion and Belief; Science and Technology; Social Issue; Sport; War, Conflict and Unrest; Weather.

Map the user's topic to the closest category ("stock market crash" -> Economy, Business and Finance). Skip filters the user did not imply.

## Using results

- Each result has a title, URL, publish date, and text excerpt.
- Cite the article URLs in your answer.
- If results are empty, retry with `days: 30` or `allow_all_dates: true` and fewer filters before reporting nothing found.
```

## Verify

Ask your agent a news question:

> What happened this week with EU AI regulation? Search Webz news.

The agent should call `news_search_by_webz` and answer with cited articles.