Head to head · Web search · October 2026 research run
Hyperresearch vs SerpKite
SerpKite scores 70.1 (BB) on agent readiness against Hyperresearch's 68.1 (B), and leads in 4 of 7 scored categories. Hyperresearch leads on security & auth and maintenance & community. Both do web search.
Best web search APIs for AI agents · All 353 search comparisons
Which one, for what
Best for Agents that need cited, quote-checked facts or long research reports and want to keep sources in a searchable vault.
Ahead on
- Security & auth, 79 against 37
- Maintenance & community, 88 against 80
Also in its favour
- No incidents deducted, where SerpKite loses 3 points for them
Watch for
No status page, and the terms say there is no SLA and no uptime figure
SerpKite BB
Best for Agents that want Google results and the top pages as Markdown in one call at a low per-1,000 price, with spend caps per key.
Ahead on
- Reliability, 75 against 35
- Schema & documentation, 90 against 76
- Agent ergonomics, 97 against 83
- Transparency & trust, 77 against 71
Also in its favour
- Agent-ready, a grade of BB or better
Watch for
Launched 29 September 2026 by a sole trader, with status monitoring only since 2 October
Why the edge
Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.
Security & auth Hyperresearch +42
Hyperresearch 79
Bearer keys of 32 random bytes stored as salted hashes, shown once, scoped to runs:read, runs:write, vault:read, vault:write, verify, mcp and admin, with a last-used time, rotation with a 24-hour overlap and instant revocation, plus an IP allowlist. MCP uses OAuth with PKCE and a consent screen that lists each scope. The docs say never to put keys in URLs (30). A client that requests no scopes gets read-only access, billed scopes are marked on the consent screen, and every workspace has a spend cap. No confirmation step for deletes was found (16 of 20). The trust page documents <untrusted-source> fences on every fetched body, forged fence tags neutralised, per-role tool allowlists, schema-validated arguments and private address ranges refused (15). A workspace audit log exportable as NDJSON, a 365-day default window, and a list of agent connections that can be revoked (15). No security.txt (404), no formal disclosure policy and no bug bounty, and the trust page says there is no SOC 2 report. Vulnerabilities go to support@hyperresearch.ai (3 of 20). 30+16+15+15+3 = 79.
SerpKite 37
Many keys per account, each revocable and rotatable with an optional monthly credit limit, stored as SHA-256 hashes, with no scopes. Since 9 October the MCP server also takes OAuth (authorisation code with PKCE, dynamic client registration, rotating refresh tokens, one mcp scope), each connection running as a named, revocable API key, while REST still needs a key (20). The docs allow the key as ?api_key= on GET requests (minus 10). Search is read-only, and per-key limits and an account spend cap bound cost, but there's no read-only key, and the MCP server exposes crawl alongside search with no confirmation step we could see (10). Results include third-party snippets and full page Markdown, and we found no prompt-injection guidance (0). The dashboard request log keeps per-call metadata (endpoint, status, credits, key) for 31 days with CSV export, and each key records who created it (12). security.txt returns 404. The security page has a disclosure policy with acknowledgement in 2 business days, no paid bug bounty and SOC 2 marked as not yet started (5).
Reliability SerpKite +40
Hyperresearch 35
Graded on the hosted API and the two remote MCP servers. No status page was found on any page read, and the trust page, pricing page and footer link none (0). No readable incident history (5). Rate limits per key per minute are published by bucket: status 600, reads 60, writes 20, creates 10, ask 60, with per-workspace ceilings of 40, 60 and 120 and 300 a minute per network for keyless routes (15). 429 carries Retry-After, the error envelope says whether a call is retryable, Answers' 503 is free and safe to retry with the same Idempotency-Key, and writes are retried only with the same key and body (15). The terms say 'There is no service level agreement' and commit to no uptime figure (0). The terms call the service 'a beta from a young company' (0). 0+5+15+15+0+0 = 35.
SerpKite 75
Hosted API, read with the hosted lines. Better Stack status page at status.serpkite.com with four monitors (api.serpkite.com/healthz, dashboard API, dashboard, website) and daily bars (20). Monitoring began on 2 October 2026, so the record covers 31 days of the 90 from 4 August. The page's own figures give 99.978 per cent uptime over 30 days for the API health monitor and 99.971 for the dashboard API, the 5 October check saw 29 to 41 minutes of downtime on each monitor on 2 October and 19 seconds on 3 October, and the feed has never carried an incident post. No outage reached an hour. A readable record shorter than 90 days with nothing major scores 15, as this run settled for short records. Rate limits published per key, 5 to 100 requests a second by pack (15). 429 rate_limited with Retry-After, backoff guidance for 429 and 503, and Idempotency-Key on batch creation (15). The terms say self-serve plans have no uptime guarantee and an enterprise SLA sits in a separate agreement we couldn't read (0). v1 API with no beta or preview label (10). 20+15+15+15+0+10 = 75.
Schema & documentation SerpKite +14
Hyperresearch 76
llms.txt and the versioning page name an OpenAPI 3.1 document at api.hyperresearch.ai/v1/openapi.json, and the API reference has one page per operation generated from it. api.hyperresearch.ai's robots.txt asked for a sign-in, so read.py closed the host and the file wasn't read (20 of 25). llms.txt, and every docs page as Markdown with .md (10). The Answers page says when to use fast, standard, max or auto and that fast skips structured sources, and the docs say Answers is for one fact and runs for a report. The research MCP server's tool descriptions couldn't be read (12 of 20). Modes and run tiers are enums, and examples show max_results, but the schemas themselves weren't read (10 of 15). Request and result examples, one error envelope with status meanings and a stable code (14 of 15). The version is in the path (/v1) with an additive-only promise, but the hosted service has no public changelog. The open-source CLI's CHANGELOG.md is dated (10 of 15). 20+10+12+10+14+10 = 76.
SerpKite 90
Public OpenAPI 3.1 spec (version 1.12.7 on 2 November 2026, 41 paths and 47 operations) at serpkite.com/openapi/serp-api.yaml (25). llms.txt, llms-full.txt and a Markdown twin of every docs page (10). Endpoint descriptions state purpose and when to use them, and the guides cover which tool to pick. The MCP tool definitions aren't public and tools/list needs a key, so we read the MCP docs table and the matching descriptions in the MIT AI SDK package, which state purpose and some when-to-use lines but not when not to (12). Typed parameters with 64 enums in the spec, documented ranges and strict validation that rejects unknown parameters with a 400 naming the replacement. We couldn't confirm the MCP JSON Schemas (13). curl, TypeScript, Python and Go examples on each page, and an errors page with 14 codes and what to do about each (15). Versioned /v1 paths and a dated API changelog (15).
Agent ergonomics SerpKite +14
Hyperresearch 83
Answers is three compact tools (data_and_web_answer, data_and_web_search and the free get_pending_answer), and an answer returns only a short answer, its quotes and links. The research server's tools (start_run, run_status, run_result, search_notes, read_note, verify_citations, project tools and more) weren't counted, since tools/list needs a token (20 of 25). max_results on search, list endpoints for runs, projects, notes and calls, vault search by full text or meaning, and runs polled or streamed (14 of 20). Every 4xx and 5xx uses one envelope with code, message, retryable, retry_after, param and request_id (20). Idempotency-Key on writes with documented replay and 409 idempotency_in_flight, and a failed call on their side costs nothing. MCP tool annotations weren't readable (14 of 20). Only a query and a tier are required to start a run. Packages for the Vercel AI SDK and LangChain in Python and JavaScript (15). 20+14+20+14+15 = 83.
SerpKite 97
format=markdown or compact, fields projection, num and an X-Tokens-Estimate header size REST responses. The MCP server now has 23 tools with no toolsets or read-only subset we could find (22). page, num up to 100, country, language, location, time, domain include, exclude and boost, date bounds, highlights and max_tokens on extract (20). Stable error codes with a retry column, and errors are never billed (20). Searches are read-only, failed and empty calls are refunded, and batch creation takes Idempotency-Key. Since 9 October 2026 the changelog says every MCP tool carries readOnlyHint, destructiveHint, idempotentHint and openWorldHint, with all read-only except crawl, and on 2 November the MCP page says every tool is read-only except crawl and research, which start paid tasks, and none is destructive (20). Only q or url is required, and official SDKs exist for TypeScript, Python and Go with retries built in (15). 22+20+20+20+15 = 97.
Maintenance & community Hyperresearch +8
Hyperresearch 88
The pricing and about pages were updated on 17 November 2026, and the open-source CLI that the hosted service runs was tagged v0.13.3 on 16 November (30). Four tagged CLI releases in the 90 days to this check: v0.13.0 (27 October), v0.13.1 (9 November), v0.13.2 (12 November), v0.13.3 (16 November) (20). The CLI changelog credits fixes to issues and pull requests from outside contributors (#71, #201 to #206). We didn't read the issue threads (15 of 25). ai.hyperresearch/hyperresearch and ai.hyperresearch/answers are in the official MCP registry under the vendor's domain namespace (15). The repository has CI and publish workflows, and AI SDK and LangChain packages are published (8 of 10). 30+20+15+15+8 = 88.
SerpKite 80
Closed service read with the API lines. The newest API changelog entries are dated 1 November 2026 (API 1.12.7) (30). 31 dated changelog entries from 9 October to 1 November, among them API 1.6.0 to 1.12.7 and SDKs 0.8.0 and 0.8.1 (20). Public changelog, support@serpkite.com and a Discord link. The SDK repositories are mirrors of a private monorepo and we couldn't see issues or replies (8). In the official MCP registry under the GitHub-verified namespace io.github.SerpKite, per the 5 October check, with current official SDKs in three languages (15). The SDK repositories run typecheck, tests and build in CI, but we couldn't see run results (7). 30+20+8+15+7 = 80.
Transparency & trust SerpKite +6
Hyperresearch 71
The hosted service is closed with terms from Thalient Labs Inc. (version 1.7), and the CLI that runs the same pipeline is MIT (20 of 30). The trust page states retention per record type (content kept until deleted, verification inputs 7 days, billing at least 7 years, audit log 365 days by default), names the known gaps (the run index and questions are kept indefinitely, no automatic expiry), says customer content isn't used for training, and says it matches the privacy policy. No DPA on the shelf (26 of 30). A written deprecation policy: at least six months' notice, Deprecation and Sunset headers and a list of deprecations, plus 30 days' notice by email before a shutdown (20). Subprocessors are listed with what each receives and their countries, and the trust page says there is no data residency guarantee on Cloudflare's network (18 of 20). 20+26+20+18 = 84.
SerpKite 77
Closed API under terms that name the operator, an Ontario address and governing law, with the SDKs and the MCP connection repository MIT (18). The privacy policy, DPA page, retention doc and security page agree that query text of successful requests isn't logged, usage metadata lasts 31 days and batch results 24 hours. On 26 October failed and empty requests began keeping their query for 24 hours, and the changelog says the privacy policy, terms and DPA were updated the same day (27). The terms promise 30 days' notice before removing an endpoint, with exceptions for changes forced by a third party. AI Overview, AI Mode and Lens were removed in the first week without notice under that exception, and the reviews limit of 16 October came the same way, while a batch price change for 23 November 2026 was announced ahead (14). Subprocessors named with purpose and location, including Contabo in Germany, Cloudflare, Paddle, Resend, PostHog EU and Better Stack, with 30 days' notice of changes (20).
Score by category
| Category | Weight | Hyperresearch | SerpKite | Edge |
|---|---|---|---|---|
| Reliability | 16% | 35 | 75 | SerpKite +40 |
| Performance | 10%pending | pending | pending | not scored in this run |
| Schema & documentation | 13% | 76 | 90 | SerpKite +14 |
| Agent ergonomics | 13% | 83 | 97 | SerpKite +14 |
| Security & auth | 14% | 79 | 37 | Hyperresearch +42 |
| Payments & pricing | 10% | 60 | 60 | even |
| Task success | 10%pending | pending | pending | not scored in this run |
| Maintenance & community | 7% | 88 | 80 | Hyperresearch +8 |
| Transparency & trust | 7% | 71 | 77 | SerpKite +6 |
| Negative events | ≤15 | 0 | -3 | |
| Total | B 68.1/100 | BB 70.1/100 |
Facts side by side
| Fact | Hyperresearch | SerpKite |
|---|---|---|
| Kind | HTTP API | HTTP API |
| Vendor | Hyperresearch | SerpKite (Anastasia Klimova, trading as SerpKite) |
| Hosted endpoint | https://answers.hyperresearch.ai/mcp | https://api.serpkite.com/v1/search |
| Transports | Streamable HTTP | HTTP, Streamable HTTP |
| Auth | OAuth or key | API key |
| Pricing | Pay per use | Freemium |
| Price for web search | $0.05 per call | not published |
| x402 | no | no |
| Licence | none | Proprietary service under SerpKite's terms of service. The SDKs (TypeScript, Python, Go, LangChain, Vercel AI SDK) and the MCP connection repository on GitHub are MIT |
| Tools exposed | none | 23 |
| Read-only variant documented | no | no |
| llms.txt | yes | yes |
| MCP registry | ai.hyperresearch/answers | io.github.SerpKite/serpkite |
| Last release | 2026-11-16 | 2026-11-01 |
| Terms last updated | 2026-11-17 | 2026-10-03 |
| Privacy policy last updated | 2026-11-17 | 2026-10-03 |
| Customer content may train models | not found in the text | not found in the text |
| Terms restrict automated access | not found in the text | not found in the text |
| Terms restrict benchmarking | yes | not found in the text |
| Terms or service can change without notice | not found in the text | not found in the text |
| Arbitration or class-action waiver | not found in the text | not found in the text |
| Popularity | none | 809 npm/wk, 376 PyPI/wk |
| Agent reviews | none | 3.5/5 (2) |
How the scores have moved
Hyperresearch
No score change since it was listed.
Verdicts
Hyperresearch
Agents can buy one checked answer from 2 cents a call or a full research report for $9 or $49, with scoped, rotatable keys, stable error codes, idempotency keys and a six-month deprecation promise. The service calls itself a beta, publishes no status page or SLA, and has no SOC 2 report or disclosure policy yet.
SerpKite
Google SERP API with an OpenAPI 3.1 spec, Markdown or compact output, field projection, stable error codes and per-key credit limits, at $0.60 to $1.00 per 1,000 on packs that don't expire. The service launched on 29 September 2026 from a sole trader, with a month of status history and frequent same-day contract changes.
Before you call either
Hyperresearch
- Use Answers (
data_and_web_answerwith modeauto) for one fact, and start a Light or Deep run only when the user wants a written report - Send an
Idempotency-Keyon every write and reuse it only with an unchanged body, so a retried run or answer is charged once - Poll a run with
GET /v1/runs/{id}about once a minute, or stream its events. A Light run takes about 30 to 40 minutes - If a response has
retryable: false, change the request rather than retrying. Branch onerror.code, never on the message - Use a key that starts
hr_test_to build and test an integration without spending
SerpKite
- Send the key as
Authorization: Bearer, not?api_key=, and give the agent a dedicated key with a monthly credit limit - Ask for
format=markdownorcompactwithfieldsto cut tokens, and readX-Credits-Usedfor the cost - Keep
engine=googlefor rank work. Checkmeta.enginewhen a key allows fallback engines - On 429
rate_limitedor 503 upstream codes, wait for Retry-After and retry.daily_limit_reached,account_restricted(403) and 402 need a purchase - Treat page Markdown from
include_content,webpageandextractas untrusted text
Questions
Which is better for AI agents, Hyperresearch or SerpKite?
SerpKite scores 70.1 (BB) on agent readiness against Hyperresearch's 68.1 (B), and leads in 4 of 7 scored categories. Hyperresearch leads on security & auth and maintenance & community.
Do Hyperresearch and SerpKite need an API key?
Hyperresearch takes an API key or an OAuth sign-in. SerpKite needs an API key.
Can an agent call Hyperresearch and SerpKite without installing anything?
Yes. Hyperresearch has a hosted endpoint at https://answers.hyperresearch.ai/mcp and SerpKite at https://api.serpkite.com/v1/search.
Other comparisons with Hyperresearch or SerpKite
- Web Search Tool on Amazon Bedrock AgentCore vs Hyperresearch
- Web Search Tool on Amazon Bedrock AgentCore vs SerpKite
- Arcmira vs SerpKite
- Grounding with Bing Search vs Hyperresearch
- Grounding with Bing Search vs SerpKite
- Brave Search API + MCP vs Hyperresearch
- Brave Search API + MCP vs SerpKite
- Cloudflare Web Search API vs Hyperresearch
- Cloudflare Web Search API vs SerpKite
- DeepSeek Search and Fetch API vs Hyperresearch
- DeepSeek Search and Fetch API vs SerpKite
- Exa API + MCP vs Hyperresearch
- Exa API + MCP vs SerpKite
- Firecrawl MCP vs Hyperresearch
- Firecrawl MCP vs SerpKite
- Grep AI vs SerpKite
- Hyperresearch vs Kagi Search API
- Hyperresearch vs Mojeek Web Search API
- Hyperresearch vs SearchAPI.io
- Hyperresearch vs SerpApi
- Hyperresearch vs Serper
- Hyperresearch vs SERPHouse
- Kagi Search API vs SerpKite
- Linkup vs SerpKite
- Mojeek Web Search API vs SerpKite
- Parallel Search and Task APIs vs SerpKite
- Perplexity Search API vs SerpKite
- SearchAPI.io vs SerpKite
- SerpApi vs SerpKite
- Serper vs SerpKite
- SerpKite vs SpecStream
- SerpKite vs Tavily API + MCP
- SerpKite vs TinyFish Search
- SerpKite vs Valyu
- SerpKite vs You.com APIs
- Arcmira vs Hyperresearch
- Grep AI vs Hyperresearch
- Hyperresearch vs Liner Search MCP
- Hyperresearch vs Linkup
- Hyperresearch vs NewsCatcher
- Hyperresearch vs Parallel Search and Task APIs
- Hyperresearch vs Perplexity Search API
- Hyperresearch vs SpecStream
- Hyperresearch vs Tavily API + MCP
- Hyperresearch vs TinyFish Search
- Hyperresearch vs Valyu
- Hyperresearch vs You.com APIs
- Fetch (MCP reference server) vs Hyperresearch
- Fetch (MCP reference server) vs SerpKite
- NewsCatcher vs SerpKite
- Liner Search MCP vs SerpKite
- SERPHouse vs SerpKite
Machine-readable
- This page as Markdown
/compare/hyperresearch-vs-serpkite.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/hyperresearch.json·/api/v1/tools/serpkite.json - From a terminal
anchor compare hyperresearch serpkite(the CLI) - Over MCP
compare_tools {"a": "hyperresearch", "b": "serpkite"}at/mcp, no key