Search1API

Search1API MCP Server

中文文档 | Deutsch

The official MCP server for Search1API — web search, news, page retrieval, sitemap discovery, and trending topics in one API.

Authentication

• OAuth-aware clients can connect to the Remote MCP URL directly, then sign in and approve access in the browser.
• Existing integrations can continue to use an API key from the Search1API dashboard.
• Every MCP request — including tool discovery (initialize, tools/list) — requires a credential. Unauthenticated requests draw the OAuth challenge, which is how clients trigger sign-in; pre-connect inspection is served by the static server card instead.

Quick Start (Remote MCP)

No installation required. Configure your MCP client with the remote URL. Use OAuth when the client supports it, or provide an API key.

Authentication

Three methods are supported — use whichever your client supports:

Prefer OAuth or the Authorization header. Query-parameter credentials can be exposed in URLs, logs, and shell history.

Optional search and news UI

Remote MCP exposes render_results at the standard endpoint above. Data tools such as search, news and ask return results without automatically showing cards. When a source list helps, the model can select or combine existing results and pass them to render_results. It returns source cards in MCP Apps clients and a complete readable text list elsewhere. Rendering does not repeat a search or consume search credits; no UI URL parameter is required.

Both HTTP and stdio advertise the square Search1API logo through standard serverInfo.icons metadata. The 512×512 PNG is served publicly at https://mcp.search1api.com/assets/icon-square.png, without authentication or redirects. The server card and registry manifest use the same icon. Clients choose whether to display it; this does not replace an icon stored separately by a plugin directory.

See the rendering contract for arguments, limits and fallback behavior. The npm package includes this renderer from version 0.8.0. Update existing installations and restart the MCP connection; npx -y search1api-mcp@latest starts the latest published package.

After changing the URL, reconnect or refresh the connector's tool definitions; clients may cache them. This URL setting applies to Remote MCP only; local stdio behavior is unchanged. See MCP Apps UI for details and release status.

Claude Desktop

{

"mcpServers": { "search1api": { "url": "https://mcp.search1api.com/mcp", "headers": { "Authorization": "Bearer YOUR_SEARCH1API_KEY" } } } }

Claude.ai (Web)

Install Search1API from the Claude directory, click Connect, and sign in with your Search1API account.

Avoid adding https://mcp.search1api.com/mcp through Add custom connector: that creates a separate custom connector, which Claude names after the URL's host (mcp.search1api.com) unless you type a name.

Cursor

Install as a Cursor plugin (recommended): this repo includes Agent Plugins plugin.json + mcp.json (portable) and .cursor-plugin/plugin.json (Cursor Marketplace metadata / logo) for Remote MCP with OAuth. Submit or install from cursor.directory / the Cursor Marketplace, then sign in when prompted.

For local testing, copy the plugin files into ~/.cursor/plugins/local/search1api (plugin.json, .cursor-plugin/, mcp.json, assets/). Do not symlink from outside that directory — Cursor rejects external symlink targets.

Or configure manually:

{

"mcpServers": { "search1api": { "url": "https://mcp.search1api.com/mcp", "headers": { "Authorization": "Bearer YOUR_SEARCH1API_KEY" } } } }

Antigravity

Install as an Antigravity plugin: antigravity-plugin/ holds the manifest, mcp_config.json (Remote MCP with OAuth), the Search1API skill, and a Marketplace README. Install it straight from GitHub, then sign in when the browser opens:

agy plugin install https://github.com/superagents-lab/search1api-mcp/tree/main/antigravity-plugin

For local testing, run agy plugin validate --strict ./antigravity-plugin and agy plugin install ./antigravity-plugin from a checkout.

VS Code

{

"servers": { "search1api": { "type": "http", "url": "https://mcp.search1api.com/mcp", "headers": { "Authorization": "Bearer YOUR_SEARCH1API_KEY" } } } }

Claude Code

claude mcp add --transport http search1api https://mcp.search1api.com/mcp \

--header "Authorization: Bearer YOUR_SEARCH1API_KEY"

Windsurf

{

"mcpServers": { "search1api": { "serverUrl": "https://mcp.search1api.com/mcp?apiKey=YOUR_SEARCH1API_KEY" } } }

Agent Skill

The Agent Skill has moved to search1api-cli. Install it with:

npm install -g search1api-cli

npx skills add superagents-lab/search1api-cli

Local Mode (stdio)

If you prefer to run the server locally, use Node.js 20 or newer with npx — no cloning required:

{

"mcpServers": { "search1api": { "command": "npx", "args": ["-y", "search1api-mcp"], "env": { "SEARCH1API_KEY": "YOUR_SEARCH1API_KEY" } } } }

For self-hosted HTTP deployments behind a proxy, add any internal hostnames that reach the Node.js process to the comma-separated MCP_ALLOWED_HOSTS environment variable. mcp.search1api.com and localhost addresses are allowed by default. Browser-based clients that send an Origin header must also have their trusted origin hostnames added to the comma-separated MCP_ALLOWED_ORIGINS variable. Requests from server-side MCP clients normally omit Origin and do not require an entry.

Tools

search

Search the web using Search1API. Results include a citable id/title/url structure. Pass a result URL to crawl when you need the full page.

Results carry published_date in metadata when the engine exposes one; github results add kind, stars, language, and num_comments.

news

Search for news articles. hackernews results link to the discussion thread and add story_url, points, and num_comments to metadata.

crawl

Extract content from a URL.

sitemap

Get all related links from a URL.

trending

Get trending topics from popular platforms.

ask

Agentic search: describe the need in natural language and Search1API picks up to five engines and a time window, drops off-topic results, and returns at most 10 citable results ranked by relevance. The result also carries intent (keywords, engines, and window used) and errors (engines that failed while others completed). Costs 5 credits per completed request; use search when you already know the engine.

maps

Find local businesses and places on Google Maps. Returns one page of up to 20 places as citable results: title is the place name, url its Google Maps link, and metadata holds the address, phone, website, category, rating, review count, coordinates, opening hours, open state, and price range when available. Costs 3 credits per page, however many places it holds.

deepcrawl

Read many pages of one site in a single task, such as a documentation section. Finds the site's URLs through its sitemap, keeps those that match the path rules, and returns up to max_pages pages as citable results: title, url, and the page's Markdown in text. The result also carries stats (discovered, matched, skipped, crawled, failed), failed (URLs that could not be read, with the reason), the API message, and result_url, a JSON file with every page in full. Costs 20 credits per started task, however many pages it reads.

The tool waits up to about 45 seconds. If the crawl is still running, it returns status: "running" with a task_id; call deepcrawl again with that task_id to collect the pages, which is free and starts no new crawl. Page text is capped at 8,000 characters per page and 80,000 in total; shortened pages have metadata.truncated: true, and result_url holds the full text.

| include_paths | No | - | Up to 20 URL path rules to keep. Without * a rule is a path prefix on segment boundaries (/docs matches /docs/api, not /docsearch); wit