How to use the Exa web search agent skill (cli-anything-exa)

What an agent skill is and what this one makes the agent do

An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to use a tool. This skill describes the cli-anything-exa command-line interface, which the agent can run to search the web and retrieve page content.

Once installed, the agent can search the web in several modes (auto, fast, instant, deep and deep-reasoning), filter results by category, date or domain, and fetch the full text or key highlights from specific pages. Results come back as machine-readable JSON.

• Web search with depth settings from fast lookups to multi-step deep research
• Filters for category (company, people, research-paper, news, personal-site, financial-report), publish dates and included or excluded domains
• Page retrieval as full text, highlights, or a summary
• JSON output for use in scripts and pipelines

Install the cli-anything-exa skill

You need Python 3.10 or newer. Download the skill ZIP from this page and unzip it into one of these locations: ~/.claude/skills/ for personal use, or .claude/skills/ inside a project folder if only that project should use it.

Alternatively, in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

The skill also depends on the command-line package. Install it from the HKUDS/CLI-Anything repository with the command below, then confirm the tool is present with the connectivity check in the next section.

• Requirements: Python 3.10 or newer
• Skill files: download the ZIP from this page and unzip into ~/.claude/skills/ or a project's .claude/skills/
• Or upload the skill in the Claude apps under Settings, Capabilities, Skills

Set your Exa API key and check the connection

Every query needs an Exa API key. Create one at https://dashboard.exa.ai/api-keys, then set it as the EXA_API_KEY environment variable. Replace the placeholder with your own key.

Run the status command to confirm the key works and the service is reachable. If it fails, the key is the first thing to check.

• The key is read from the EXA_API_KEY environment variable
• Keep the key out of shared files and screenshots

Ask the agent for searches in plain language

You do not need to type commands yourself. Describe the task in a normal request that involves finding web pages, papers, news or company information, or reading a specific URL. The agent can then choose the search type, category and filters that fit the request.

If you want to control the settings, name them in your request, for example asking for a deep search, only news from a date range, or results limited to certain domains. The skill's own search options are listed in the command reference below.

• Search for recent news on a topic and only include results from a given date onward
• Find academic papers on a subject, returning around 20 results
• Look up funding history or background on a company
• Read the full text of a specific article URL and summarise it

Command examples the agent can run

These are the patterns the skill documents. The --json flag at the front returns machine-readable output. Use --type fast for quick keyword lookups and --type deep for research that synthesises across many sources.

Fetching a page uses the contents command with a URL. Search options include --num-results (1 to 100, default 10), --content (highlights, text, summary or none), --freshness (smart, always or never), --include-domains and --exclude-domains, --from and --to for ISO 8601 dates, and --location for a two-letter country code.

• Fast keyword lookup restricted to a site, using highlights
• Deep research on a topic, using full text
• Academic paper discovery in the research-paper category
• Company intelligence in the company category
• News monitoring from a start date
• Fetching full article text for summarisation

Safety notes: scripts, network and data

This skill runs scripts and uses the network. Every search term and every URL the agent fetches is sent to Exa's external service, so do not use it for sensitive or confidential research.

The skill also includes an interactive mode. Running cli-anything-exa with no subcommand opens a prompt where you type commands without the prefix, and you type exit or quit to leave. Each query has a cost, which the source reports in the cost_dollars field of JSON output.

• Search terms and fetched URLs leave your machine and go to Exa
• Each query incurs a cost; check cost_dollars in JSON output
• Category company and people do not support date or domain-exclude filters
• The skill needs internet access and does not work offline

Tips for better results and lower cost

Prefer highlights over full text in agent pipelines. The source describes highlights as 10 times more token-efficient than text, so reserve text for pages where you need the complete content.

Use deep search only when you need synthesis across sources, since it is slower. Narrow results with --category or domain filters before increasing --num-results.

• Start with the default auto type, and move to deep only for research questions
• Use --include-domains to keep results to sources you trust
• Remember that company and people categories ignore date and domain-exclude filters