Web search APIs and MCP servers for AI agents
Search engines and answer APIs that return fresh web results, snippets or full page content in agent-friendly formats.
14 web search APIs an AI agent can call: auth method and header, official MCP server (11), free tier (9), docs links. Machine-readable JSON included. Agents: fetch catalog/web-search.json instead of parsing this page. Last checked 2026-09-24.
Ratings and usage (★) fetched: .
| Service | What an agent can do | Auth | MCP server | Free tier | Details (with sources) | Notes | Proof (ratings, usage) |
|---|---|---|---|---|---|---|---|
| Brave Search API | Lets an agent query Brave's independent web/image/video/news search index and get cited, LLM-ready results. | API keyX-Subscription-Token: <API_KEY> | official | $5 free monthly credits (~1,000 queries on Search plan); credit card required at signup as anti-fraud measure | Base URL: https://api.search.brave.com/res/v1 Free plan: $5 in credits (~1,000 Search requests at $5/1,000), per month, card required ↗ Rate limits: Search plan 50 queries/sec; Answers plan 2 queries/sec ↗ Data policy: Zero Data Retention offered only on the Enterprise plan; storage rights require special plans ↗ Example call: run a web search: GET https://api.search.brave.com/res/v1/web/search?q={query} ↗ Unit price: Search plan: $5 per 1,000 requests ($0.005 per query); $5 free monthly credits ↗ MCP tools: 8: brave_web_search, brave_local_search, brave_video_search, brave_image_search, brave_news_search, brave_summarizer … ↗ | Official MCP server (@brave/brave-search-mcp-server) is self-hosted only (npx/Docker); STDIO is the default transport since v2.x, HTTP optional; no vendor-hosted endpoint. Answers plan is separate ($4/1,000 req + token costs). | GitHub 1,467★ brave/brave-search-mcp-server Smithery 119.0k uses SourceForge 4.5★/2 reviews |
| Exa | Lets an agent run neural or keyword web search, retrieve page contents, and find similar links via Exa's API. | API keyAuthorization: Bearer <EXA_API_KEY> (or x-api-key header) | official | free tier: $10 free credits/month (~1,400 searches), no payment method required | Base URL: https://api.exa.ai Free plan: $10 in credits (~1,400 searches), balance resets to $10 on the 1st of each month, no card ↗ Rate limits: /search, /answer 10 QPS; deep search 5 QPS; /contents 100 QPS; /websets 20 QPS; agent runs 5 QPS and 50 active runs ↗ Data policy: Zero Data Retention available as an Enterprise (custom plan) feature ↗ Example call: search the web: POST https://api.exa.ai/search ↗ Unit price: Search: $7 per 1,000 requests (up to 10 results; $0.007 per search), +$1 per 1,000 extra results; deep search $12-$15 per 1,000; contents $1 per 1,000 pages per content type ↗ MCP tools: 4: web_search_exa, web_fetch_exa, web_search_advanced_exa, agent_run ↗ | Remote MCP works keyless (rate-limited); for your own plan send x-api-key or sign in via OAuth at https://mcp.exa.ai/mcp?login (agent_run requires auth). Self-hosted npm package (exa-mcp-server) also available. | SourceForge 5.0★/1 review |
| Jina AI Search / Reader | Search the web or convert any URL into clean LLM-ready markdown via simple HTTP GET requests. | API keyAuthorization: Bearer <JINA_API_KEY> (works without a key too, but at a lower rate limit) | official | No key needed for Reader only: 20 RPM (Search is blocked without a key); free API key gives Reader 500 RPM / Search 100 RPM plus 10M free tokens one-time on signup | Base URL: https://r.jina.ai Free plan: 10M free tokens per new API key; Reader also usable without key at 20 RPM, one-time ↗ Rate limits: No key: Reader 20 RPM, Search blocked; free/paid key: Reader 500 RPM, Search 100 RPM; premium: 5,000/1,000 RPM ↗ Data policy: API requests, inputs and outputs never used for training; Reader caches content 5 minutes (X-No-Cache to bypass) ↗ Example call: convert a URL to LLM-ready markdown (Reader): GET https://r.jina.ai/{url} ↗ Unit price: Tokens shared across Jina APIs: $50 per 1B tokens ($0.05 per 1M); Reader bills the output tokens of each page, Search costs a fixed 10,000+ tokens per request ↗ MCP tools: 12: primer, read_url, capture_screenshot_url, guess_datetime_url, search_web, search_arxiv … ↗ | Reader is usable without an API key at a low rate limit; Search needs a key. Interactive docs/playground at https://api.jina.ai/scalar. Official remote MCP server at https://mcp.jina.ai/v1 (github.com/jina-ai/MCP). | — |
| Kagi Search API | Run programmatic, ad-free premium web/news/image/video searches and extract page markdown via a paid REST API. | API keyAuthorization: Bearer <API_KEY> (v1 API; legacy v0 used 'Bot <token>') | official | no free tier: pay-per-use, $12 per 1,000 search requests, invoiced monthly or at $100 usage, whichever first | Base URL: https://kagi.com/api/v1 Free plan: none ↗ MCP tools: 2: kagi_search_fetch, kagi_extract ↗ | Requires a funded/billed Kagi account; on Family/Team plans only the account owner can access the API portal. Official hosted MCP endpoint also available at https://mcp.kagi.com/mcp using just an API key. | GitHub 528★ kagisearch/kagimcp |
| Linkup | Lets an agent search, fetch pages, or run deep multi-source research via Linkup's Search/Fetch/Research API endpoints. | API keyAuthorization: Bearer <LINKUP_API_KEY> | official | $20 free credits/month for new accounts signing up with a professional email | Base URL: https://api.linkup.so/v1 Free plan: $20 credits, topped back up to $20 each month (professional-email signups) ↗ Async jobs: research tasks are started then polled for status/results (MCP tools linkup-research + linkup-get-research) ↗ Example call: web search with a sourced answer: POST https://api.linkup.so/v1/search ↗ Unit price: Search: $0.005 per call (flash/fast/standard, searchResults), $0.006 with sourcedAnswer/structured; deep $0.05-$0.055 ↗ MCP tools: 4: linkup-search, linkup-research, linkup-get-research, linkup-fetch ↗ | Also supports the x402 protocol: pay-per-request in USDC on Base with no account required; Extract endpoint is in closed beta; MCP endpoint also supports ?apiKey= query param auth. | Smithery 213 uses |
| Mojeek Search API | Query Mojeek's independent web index for titles, URLs and snippets in JSON or XML, with country/language boosting. | API keyapi_key query parameter (https://api.mojeek.com/search?q=...&api_key=<KEY>) | — | limited-query free trial available on request; paid plans from £2 CPM (Startup, 5 req/s, 100k/day cap) | Base URL: https://api.mojeek.com Free plan: limited free trial on request (contact Mojeek), one-time ↗ Rate limits: Startup: 5 queries/sec, 100,000/day; Business: 10 queries/sec, 400,000/day; Enterprise unlimited ↗ Data policy: All plans include storage rights and AI usage permissions; results may be re-ranked and combined with other engines ↗ Example call: search the web: GET https://api.mojeek.com/search?q={query}&api_key={API_KEY}&fmt=json ↗ Unit price: Startup plan GBP 2 per 1,000 queries (GBP 0.002/query); Business GBP 3 per 1,000; excl. VAT ↗ | Sign-up is not instant self-serve, requires contacting Mojeek via their contact form to get a key/plan (Startup/Business/Enterprise). Independent (non-Google/Bing-derived) index. | — |
| Parallel Web Systems Search API | Lets an agent run Parallel's low-latency Search API for web research, content extraction, and entity discovery. | API keyx-api-key: <PARALLEL_API_KEY> | official | free: 5,000 requests/month across APIs, plus up to $80 signup credit and $5/month free credits | Base URL: https://api.parallel.ai/v1 Free plan: up to 5,000 requests/month, plus up to $80 signup and $5/month credits ↗ Rate limits: Varies by API: Task 2,000/min, Search 600/min, Find All 25/hr ↗ Data policy: Zero data retention offered on Enterprise plans ↗ Example call: run a web search: POST https://api.parallel.ai/v1/search ↗ Unit price: Search: $1 per 1,000 requests (turbo/fast modes) or $5 per 1,000 (basic/advanced; advanced is the default); +$1 per 1,000 additional results beyond 10 ↗ MCP tools: 2: web_search, web_fetch ↗ | The remote MCP server works without an account/API key at lower rate limits; an Authorization: Bearer key raises them. The main REST API requires an API key from the dashboard. Search defaults to mode advanced ($5 per 1,000 requests); set mode fast for $1 per 1,000 | — |
| Perplexity Sonar / Search API | Lets an agent call Perplexity's Search API or MCP tools (search/ask/research/reason) for real-time grounded web answers. | API key / OAuthAuthorization: Bearer <PERPLEXITY_API_KEY> (remote MCP also supports OAuth sign-in) | official | Paid only: no confirmed free tier; pay-as-you-go, Search API billed at $5 per 1,000 requests ($1 per 1,000 for the fast search variant) | Base URL: https://api.perplexity.ai Free plan: none ↗ Example call: run a web search: POST https://api.perplexity.ai/search ↗ Unit price: Search API: $5 per 1,000 requests ($0.005 per request); fast search $1 per 1,000 ↗ MCP tools: 4: perplexity_search, perplexity_ask, perplexity_research, perplexity_reason ↗ | Sonar Chat Completions endpoint is being migrated to the new Agent API with a migration deadline around 2026-09-27; platform now exposes four APIs: Search, Router, Agent, Embeddings. | — |
| SearchApi.io | Fetch structured JSON results from Google, Bing, YouTube, Amazon, TikTok and 50+ other search engines via one REST API. | API keyapi_key=<KEY> as query param, or Authorization: Bearer <KEY> | official | Trial: 100 free requests one-time on signup, no credit card required; then Developer plan $40/mo for 10,000 searches | Base URL: https://www.searchapi.io/api/v1 Free plan: 100 free requests on signup, one-time, no card ↗ Rate limits: Up to 20% of the plan's monthly credits can be used per hour ↗ Example call: run a Google web search: GET https://www.searchapi.io/api/v1/search?engine=google&q={query} ↗ Unit price: 1 successful search = 1 request; $4 per 1,000 on Developer ($40/mo for 10,000), down to $1 per 1,000 on Octo 5M; failed requests not billed ↗ | Separate OpenAPI spec per engine (e.g. /openapi/google.yaml); official hosted MCP server at https://www.searchapi.io/mcp (OAuth, or X-MCP-Token header for clients without OAuth). | — |
| SearXNG (self-hosted JSON API) | Query a self-hosted SearXNG metasearch instance's JSON API to aggregate results from many search engines at once. | nonenone by default; self-hosted instance owner controls access (can add auth/proxy in front) | community | free, no key (self-hosted; you run and pay for the instance yourself) | Free plan: no_key, no card ↗ Example call: run a metasearch query (JSON output): GET https://{instance}/search?q={query}&format=json ↗ MCP tools: 4: searxng_web_search, searxng_search_suggestions, searxng_instance_info, web_url_read ↗ | Not a vendor with a signup/API key, it's open-source software you self-host (or point at a trusted public instance from searx.space). JSON/CSV/RSS output formats must be explicitly enabled by the instance admin in settings.yml (search.formats); most public instances disable JSON and return 403. Included because it gives agents a genuine no-signup programmatic search interface once deployed. | GitHub 1,260★ ihor-sokoliuk/mcp-searxng |
| SerpApi | Lets an agent scrape structured JSON results from Google and 20+ other search engines via SerpApi's REST API. | API keyapi_key query parameter (also usable as header depending on client library) | official | free plan: 250 searches/month, 50 requests/hour throughput | Base URL: https://serpapi.com Free plan: 250 searches, per month, card required ↗ Rate limits: Free plan: 50 searches/hour throughput; cached, errored and failed searches not counted ↗ Data policy: ZeroTrace Mode (higher-tier plans): search parameters, queries and results not retained ↗ Async jobs: optional async=true submits search to retrieve later (via Search Archive); default is synchronous ↗ Example call: run a Google search: GET https://serpapi.com/search?engine=google&q={query}&api_key={API_KEY} ↗ Unit price: 1 search per request; Starter $25/mo for 1,000 searches ($0.025/search), Developer $75/mo for 5,000 ($0.015/search); cached searches free ↗ MCP tools: 3: search, search_table, search_dashboard ↗ | Cached searches are free and don't count toward the monthly search quota. Hosted MCP accepts Authorization: Bearer <key> or a /<API_KEY>/mcp path; source at github.com/serpapi/serpapi-mcp. | GitHub 173★ serpapi/serpapi-mcp |
| Serper | Lets an agent fetch fast, low-cost Google search results (web, images, news, maps, shopping) via Serper's REST API. | API keyX-API-KEY header (per widely published examples; could not independently re-confirm via automated fetch, see notes) | community | Trial: 2,500 free queries one-time on signup, no credit card required; then prepaid query packages | Free plan: 2,500 free queries, one-time, no card ↗ Unit price: 1 credit per query; prepaid packs $50 for 50k ($1.00/1k) down to $3,750 for 12.5M ($0.30/1k); credits valid 6 months; 2,500 free queries ↗ MCP tools: 2: google_search, scrape ↗ | Docs/playground are login-gated. No subscriptions: prepaid credits from $1.00/1k down to $0.30/1k, valid 6 months. No official MCP server found; listed repo is the most-starred community one. | GitHub 169★ marcopesani/mcp-server-serper |
| Tavily | Lets an agent call Tavily's search API or MCP server to run web searches and get AI-ready summarized results. | API keyAuthorization: Bearer tvly-<API_KEY> | official | free tier: 1,000 API credits/month, no credit card required | Base URL: https://api.tavily.com Free plan: 1,000 API credits, per month, no card ↗ Rate limits: Dev keys 100 RPM, production keys 1,000 RPM; Crawl 100 RPM; Research task creation 20 RPM ↗ Example call: search the web: POST https://api.tavily.com/search ↗ Unit price: Search: 1 credit (basic/fast/ultra-fast) or 2 credits (advanced); pay-as-you-go $0.008 per credit ↗ MCP tools: 4: tavily-search, tavily-extract, tavily-map, tavily-crawl ↗ | Advanced search depth costs 2 credits vs 1 for basic/fast; remote MCP: OAuth at https://mcp.tavily.com/mcp or API key via https://mcp.tavily.com/mcp/?tavilyApiKey=<key>; repo: github.com/tavily-ai/tavily-mcp. | Smithery 1,575 uses |
| You.com API | Lets an agent call You.com's Search/Answer/Research API or MCP server for LLM-ready web search and grounded answers. | API keyX-API-Key header (or YDC_API_KEY env var) | official | Free: 100 queries/day with no API key or sign-up through the MCP free profile (search and discover only); new API-key accounts also get $100 in one-time credit | Base URL: https://ydc-index.io/v1 Free plan: 100 queries/day keyless (you-search, you-discover via MCP ?profile=free); plus $100 one-time credit for API-key accounts, no card ↗ Data policy: Zero Data Retention offered as an enterprise feature ↗ Example call: run a web search: POST https://ydc-index.io/v1/search ↗ Unit price: Web Search API: $5.00 per 1,000 calls ($0.005 per search); Contents $1 per 1,000 pages; Research $12 per 1,000 calls ↗ MCP tools: 6: you-search, you-contents, you-balance, you-discover, you-research, you-finance ↗ | Current endpoint is POST https://ydc-index.io/v1/search (recommended); legacy GET still works but won't get new features; also offers a separate keyless Docs MCP server at you.com/docs/_mcp/server for searching You.com's own docs. | SourceForge 5.0★/1 review |
Other categories
- Web scraping and browser automation
- Knowledge and research data
- Image generation and editing
- Video generation and editing
- Speech and audio
- Music generation
- 3D generation and assets
- Architecture, CAD and BIM
- Diagrams and software architecture
- Design and UI
- Documents, OCR and presentations
- Translation and language
- Code execution sandboxes
- Developer platforms and DevOps
- Cloud and hosting
- Databases
- Vector search, memory and RAG
- Email, team chat and notifications
- SMS and messaging apps
- Voice agents and telephony
- Productivity and workspace
- CRM, support and marketing
- Social media
- Maps, geolocation and weather
- Finance, payments and market data
- E-commerce
- Automation and integration platforms
- Model APIs and inference