# Look something up to ground a reply

> Look something up to ground a reply: 3 companies fully do it, 19 with a limit, 58 carry the tag. Every listing in Web search APIs with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/search.ground-fact/
- Markdown: https://www.anchorterminal.com/niches/search.ground-fact/index.md (~2,250 tokens)
- Slim: https://www.anchorterminal.com/niches/search.ground-fact/index.min.md (~430 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/search.ground-fact/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Two or three companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 3 | 19 | 58 | 24 | 19 |

## What counts as done

Five to ten ranked results with title, URL and a snippet or highlight, returned in one call, with no 429 when three calls land within a second.

Measured, for our own tests: at least one result is on-topic for the query, and the call returns within 3 seconds.

How agents ask for it: search the web for X; what is the latest on X; find me pages about X; is this still true

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Linkup (Linkup Technologies SAS)**: /search with depth flash, fast, standard or deep and outputType searchResults returns name, url, snippet, content and date per result with maxResults, at 10 queries a second per organisation; an EU user is happy with it.
- **Tavily API + MCP (AlphaAI Technologies Inc. dba Tavily)**: /search with max_results, topic and country at 1 credit for basic depth, 1,000 free credits a month with no card, a keyless mode, and 100 requests a minute on the free plan; an informal test ranked it joint first.
- **You.com APIs (SuSea, Inc., d/b/a you.com)**: POST /v1/search returns url, title, description, snippets and page_age with count up to 100, freshness and country; the MCP server's profile=free answers with no key.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Web Search Tool on Amazon Bedrock AgentCore (Amazon Web Services, Inc. (regional AWS entities by account location))**: Results carry title, URL, snippet text and publishedDate with maxResults 1 to 25, at 10 requests a second by default. The first call needs an AWS account, a gateway, a target and IAM or Cognito auth, and no tool error responses are documented.
- **Brave Search API + MCP (Brave Software, Inc.)**: Web Search returns up to 20 titled, described results per call with freshness and country, but rate limits are a 1-second window per plan (the guide's example is 1 request a second) and Open WebUI users hit 429s on back-to-back queries.
- **Cloudflare Web Search API (Cloudflare, Inc.)**: One request shape (provider, limit 1 to 10, byokAlias) over Ceramic, Exa or Linkup, but credits need a payment method and two accounts got 2024 results or zero results from the default Ceramic provider.
- **DeepSeek Search and Fetch API (Hangzhou DeepSeek Artificial Intelligence Co., Ltd.)**: Cheap and working per two testers, but judged weak against Parallel and Exa, and politically sensitive queries returned zero results; DeepSeek's terms bar automated copying so its docs were not re-read.
- **Exa API + MCP (Exa Labs Inc.)**: numResults with highlights at 10 QPS on a free tier of $10 a month with no card, but the API and hosted MCP answer Cloudflare 403 from some datacenter IPs and one MCP search timed out after 300 seconds.
- **Kagi Search API (Kagi Inc.)**: POST /search with limit, page 1 to 10, region and date filters, and two users call it excellent inside agent harnesses, but there is no free tier, a payment method comes first, and no rate numbers are published.
- **Liner Search MCP (Liner)**: POST /tools/search/web returns title, url, description and date with max_results 1 to 20 and no per-second limit, but since 11 November Premium Data is on by default at $2 per 1,000 unless include_knowledge is false, and no free allowance appears in the docs.
- **Mojeek Web Search API (Mojeek Limited)**: JSON results with url, title and desc, up to 10 a request at 5 queries a second on Startup, judged close to Google by one user; but the key travels in the query string and a daily limit arrives as a status string inside a 200.
- **NewsCatcher (NewsCatcher, Inc.)**: Ranked articles with title, url, snippet (200 characters on the free plan) and score, but the index is the last 30 days of news only, so general lookups fall outside it.
- **Parallel Search and Task APIs (Parallel Web Systems Inc.)**: fast mode answers in about 700 ms at $1 per 1,000 with 600 requests a minute, and a Keenable engineer ranked it top, but search_queries is required (422 without it) and an omitted mode defaults to advanced at $5 and about 3 seconds.
- **Perplexity Search API (Perplexity AI, Inc.)**: POST /search returns title, url, snippet, date and last_updated with max_results 1 to 20 and up to five queries a request, but there is no free allowance, credits are prepaid, and rate limits start at 1 QPS on Tier 0.
- **SearchAPI.io (SearchApi, LLC)**: Google organic results as JSON with gl, hl, location, device and page, 401 and 429 as JSON errors, but after 100 free requests plans start at $40 a month and bursts are paced to 20 per cent of the monthly quota an hour (listing).
- **SerpApi (SerpApi, LLC)**: Google results as JSON or Markdown with location, gl, hl and start, and two happy users, but the key goes in the query string, the free plan is 250 searches a month, and the status page logged 35 incidents in three months (listing).
- **Serper (Serper)**: Google organic title, link, snippet and position with 2,500 free queries and no card, and two users call it the best price for the quality, but Serper publishes no reference: gl, hl, tbs and page are documented only by third parties, and there is no status page.
- **SERPHouse (Patoliya Infotech LLP)**: Google organic JSON with q, domain, lang, device and loc and a 4,000-credit trial without a card, but 60 requests a minute on every self-serve plan and a 200 can carry status error 'Please try again'.
- **SerpKite (Anastasia Klimova, trading as SerpKite)**: Google web search with country, language, location, num, page, time, device, domain and date filters, markdown or compact output, and 429 and 503 carrying Retry-After; but the vendor is a sole trader whose API opened in September 2026.
- **SpecStream (SpecStream SAS)**: GET /v1/web-search returns an answer, rows and up to 10 sources with url, title, snippet and published at $5 per 1,000 ($1 on speed=fast), but it is answer-first and a 503 when the minute's calls are used names no limit.
- **TinyFish Search (Tiny Fish, Inc.)**: Ranked title, snippet, url and date with location and language, free up to 12,000 requests a day, but 30 requests a minute per person throttles agents that fire several queries a turn and there is no result-count parameter.
- **Valyu (Valyu.Network Ltd)**: Results with title, url and content, max_num_results up to 20 and fast_mode, 2,000 free credits a month with a work email and no card; but billing is per result and the pay-as-you-go limit is 5 requests a second.

- not this job: Arcmira (Arcmira, Inc.): Indexes YouTube and podcast transcripts only; GET /v1/search returns spoken passages, not web pages.
- not this job: Grounding with Bing Search (Microsoft Corporation): Developers and end users do not get the raw content returned from Grounding with Bing; only a model response with url_citation annotations.
- not this job: Fetch (MCP reference server) (Model Context Protocol, a Series of LF Projects, LLC): A single fetch tool; no search.
- not this job: Grep AI (Parcha Labs, Inc.): POST /api/v2/runs starts a research agent (low effort takes 10 to 20 seconds); no ranked results endpoint.

## What stops agents finishing

- free tier at 1 request a second when the UI fires three queries at once (Brave in Open WebUI)
- a card needed before the first call
- daily cap of about 1,000 on the free tier with no fallback provider
- Cloudflare 403 from datacenter IPs

## Around it

After: search.fetch-url, web-scrapers.page-to-markdown.

## Where this comes from

Weight 24 of 100 in Web search APIs: 19 of 44 accounts describe plain lookups from an agent or chat UI; the most common job in both Hacker News threads and the Open WebUI and OpenClaw issues

As data: https://www.anchorterminal.com/api/v1/use-cases/search.ground-fact.json, or the MCP tool get_use_case with id search.ground-fact. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
