A web search skill for Claude Code and other terminal agents
An agent with a search tool usually fires one default query and pastes the snippets back. The Search1API skill teaches it to pick the command, the engine, the result count, and whether to read the pages at all.
Install β bash
The skill picks the command, the engine, the result count, and whether to read the pages at all.
A link in the conversation gets crawled for its text, not searched for again.
A quick lookup stays shallow; a research question widens, then reads the best few in full.
"Latest" turns into a time range on the request instead of extra query terms.
Try saying
search for the latest AI news / what does this link say? / what's trending on GitHub?
What an agent skill adds over a tool
An MCP server tells an agent what it can call. A skill tells it when and how, which is the part that decides whether the answers are any good. The Search1API skill is a set of instructions the agent reads at the point of use: it maps a URL in the conversation to a page read rather than a search, sends a quick factual lookup down a different path than a research question, and turns a word like "latest" into a time filter instead of leaving it in the query string. It drives the s1 CLI, so the agent runs ordinary commands you can read in the transcript.
What changes once it is installed
Each of these is a decision the agent makes on its own, without you naming a command or a flag.
A URL means read it
When a link appears in the conversation, the agent crawls that page for its full text instead of searching for the page it already has.
Depth matches the question
A quick factual lookup gets a handful of results and no crawling. A research question gets a wider result set, then reads the most useful three to five in full.
Recency becomes a filter
Words like "latest" or "today" become a time range on the request rather than extra terms in the query, which is what actually narrows the results.
Workflows, not one-shot calls
Deep research, summarizing a URL, and following up on a trend are defined end to end, so the agent chains several calls instead of stopping at the first response.
An answer, not a result dump
The skill has the agent synthesize what it found and cite the sources, rather than pasting raw results back into the conversation.
The CLI it runs on
s1 puts the whole API in your terminal, and because it prints JSON on demand, in your shell scripts too. The skill calls these same commands, so anything the agent does you can run by hand.
Install and log in β bash
Log in with OAuth, not an API key
s1 login is a real OAuth 2.1 flow. The CLI discovers the authorization server, registers itself as a public native client, and runs Authorization Code with PKCE. You approve it in the browser, and it refreshes its own access token from then on. An API key still works for CI and scripts where an interactive login is impractical.
Commands
s1 search "<query>"Search the web
s1 news "<query>"Search the news
s1 crawl <url>Read a page
s1 sitemap <url>List a site's links
s1 trending <github|hackernews>Trending topics
s1 balanceRemaining credits
s1 loginAuthorize with OAuth 2.1 in the browser
Examples
Everyday usage β bash
Where it earns its place
Claude Code checking current library docs before writing against an API it has not seen.
A terminal agent researching a topic across several sources and citing them back.
Shell scripts that pipe --json into jq for a scheduled report.
Summarizing a URL a teammate pasted, without leaving the terminal.
Checking what is trending on GitHub or Hacker News at the start of a session.
FAQ
What is an agent skill?
A skill is a set of instructions an agent reads when it is deciding how to use a capability. Where a tool definition says a search function exists and lists its parameters, a skill says which parameters suit which question, which workflows to chain, and how to present what comes back. The Search1API skill covers search, news, page reading, and trending.
Do I need the CLI as well?
Yes. The skill drives the s1 CLI, so the CLI has to be installed and logged in first. If it is not, the skill walks the agent and you through it on first use.
Which agents can use it?
Any agent host that supports skills and can run a shell command, which includes Claude Code and other terminal agents. The skill is published on Smithery as well, for hosts that install from there.
Is this the same as the MCP server?
No, and they solve different halves of the problem. The MCP server exposes the tools; the skill supplies the judgment about using them. If your client speaks MCP and you want tools with no local install, use the MCP server. If your agent works in a terminal and you want it to search well rather than merely be able to search, install the skill.
How much does it cost?
The skill and the CLI are free. They call the same API and spend the same credits: 1 credit for a search or news request, plus 1 for each page successfully crawled. Deep research modes crawl pages, so they cost more than a plain lookup. New accounts get 100 credits free without a card.
Can I use it in CI?
Yes. Set SEARCH1API_KEY in the environment and the CLI uses that instead of the saved OAuth tokens, which is the practical option where no browser is available. Add --json to get machine-readable output.
Explore more
Teach your agent how to search
Two commands to install, an OAuth login, and 100 free credits to try it with.