Head to head · Web search · October 2026 research run

Hyperresearch vs Kagi Search API

Hyperresearch scores 68.1 (B) on agent readiness against Kagi Search API's 57.3 (C), and leads in 5 of 7 scored categories. Kagi Search API leads on reliability. Both do web search.

Best web search APIs for AI agents · All 353 search comparisons

Which one, for what

Hyperresearch B

Best for Agents that need cited, quote-checked facts or long research reports and want to keep sources in a searchable vault.

Ahead on

  • Schema & documentation, 76 against 71
  • Agent ergonomics, 83 against 75
  • Security & auth, 79 against 64
  • Payments & pricing, 60 against 30
  • Maintenance & community, 88 against 41

Also in its favour

  • Free to start without a card

Watch for

No status page, and the terms say there is no SLA and no uptime figure

Kagi Search API C

Best for An owner who already pays for Kagi and wants its ranking, lenses and domain rules inside an agent, with page text in the same call.

Ahead on

  • Reliability, 43 against 35

Also in its favour

  • Runs on your own machine

Watch for

Public preview since 21 May 2026. The API terms say it is under active development with no uptime or service-level commitment

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.

Maintenance & community Hyperresearch +47

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.

Kagi Search API 41

The last dated API change is the Extract API fix in the changelog post of 21 August 2026, 75 days before this check. The posts of 10 September and 16 October carry no API item (20). That is the only dated API entry in the 90 days from 6 August (0). The MCP server's repository gained commits on 27 and 28 October 2026 (compact JSON results, readOnlyHint, per-page extraction errors) and a version bump to 1.1.0 with no tag, so no release is shown; PyPI didn't resolve for us. A public changelog that credits user reports, developers@kagi.com, a Discord #api forum and GitHub discussions. We could not read the issue list, since the GitHub API answered 403 (9). Four generated clients last synced on 26 May 2026, Python tagged only as 1.0.0rc3, TypeScript not on npm on 9 October, and kagimcp tagged last as v0.1.5 (8 April 2026). We did not look the server up in the MCP registry (8). The Python client has a generated CI workflow and the MCP server has a lock file but no CI workflow (4). 20 + 0 + 9 + 8 + 4 = 41.

Payments & pricing Hyperresearch +30

Hyperresearch 60

No x402, MPP or L402 (0). Per-call prices for Answers and flat per-run prices (Light $9, Deep $49) are published without login, with rerun discounts and refund rules (30). $5 of Answers credit and the first Light report free on every account, no card needed, and hr_test_ keys are never billed (30). A person signs up through Clerk in a browser and creates keys in the console, and MCP clients authorise in a browser (0). 0+30+30+0 = 60.

Kagi Search API 30

No x402, MPP or L402 (0). $12 per 1,000 searches and $4 per 1,000 extracted pages, published without a login, unchanged on 4 November (20). No free tier or trial for the API. The quick start asks for a payment method, and the $5 credit of 21 May 2026 was a one-off for subscribers (0). A person signs up in a browser and creates the key in the portal (0). 0 + 20 + 0 + 0 = 20. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 20 is 30.

Security & auth Hyperresearch +15

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.

Kagi Search API 64

Several named keys an account, each limited to chosen products and optionally to IP addresses or ranges, revoked by deleting. The v1 key travels only in the Authorization header. No expiry or OAuth (27). The API only reads. Per-key product limits and a spending cap in USD narrow what a leaked key can do (16). Search and Extract return third-party page text, and we found no prompt-injection guidance in the reference, the docs or the MCP README (2). The portal shows use count and last-used time per key and requests and cost per day per API, and every response carries a trace ID. No per-call log (7). security.txt with a contact, a bug bounty with a safe harbour policy, and an independent audit by Illumant cited on the security page, whose date we did not read. No SOC 2 or ISO 27001 statement found. security.txt re-read on 4 November (12). 27 + 16 + 2 + 7 + 12 = 64.

Reliability Kagi Search API +8

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.

Kagi Search API 43

Status page at status.kagi.com (Instatus), found through a web search on 4 November, with a Kagi API group of nine regional components such as [API] us-east4 and [API] europe-west2, and history back to 2025 (20). In the 90 days from 6 August its feed and incident list show [API] incidents on single regions only. us-central1 on 12 September (1 hour 3 minutes), 19 September, 12 October and 19 October (2 to 3 minutes each), europe-west2 on 19 October (1 minute) and 20 October (8 hours 11 minutes, the region taken out of the global load balancer), asia-east2 on 30 September (4 minutes), australia-southeast1 on 11 September (1 minute) and asia-southeast1 on 7 August (about a minute). Each hit one region of nine, so all read as partial and the record is minor incidents only (20). The six 'API outage' entries (10, 20 and 26 August, two on 14 October and 22 October, 5 to 21 minutes each) name the API component of the Kagi Assistant group, not the Search API, and the 27 September database maintenance had an announced window. No rate limit numbers are published. The terms say only that limits and quotas apply per plan (0). The OpenAPI file documents 429 as 'Rate limited or usage limit exhausted' with no Retry-After header or backoff guidance. The calls are read-only, and Kagi's own MCP server retries 429 and 5xx twice by default (3). The pricing page lists SLAs under Enterprise by quote, and the API terms say there is no uptime or service-level commitment unless agreed in writing (0). The OpenAPI file still describes a preview period (0). 20 + 20 + 0 + 3 + 0 + 0 = 43.

Agent ergonomics Hyperresearch +8

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.

Kagi Search API 75

limit caps results, extraction of page text is opt-in, snippet length follows an account setting, and an experimental Markdown response format exists. The MCP server has two tools (22). page 1 to 10, limit 1 to 1,024, lens and filters for domains, keywords, file type, dates and region (20). Errors return a namespaced code, a message, the field location and a trace ID, but no code catalogue is published and the linked error pages are not live (12). Both operations are read-only, so a retry is safe, though the docs do not say so and failed-call billing is stated only for extraction. The released MCP server sets no read-only annotations. A commit of 27 October adds readOnlyHint to its source with an untagged 1.1.0 version bump, not yet shown as released (10). Only query is required and defaults are stated. Generated clients exist for Python, TypeScript, Go and Rust, installable from GitHub. @kagi/api returned 404 on npm on 9 October (11). 22 + 20 + 12 + 10 + 11 = 75.

Schema & documentation Hyperresearch +5

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.

Kagi Search API 71

Public OpenAPI 3.0 file at kagi.com/api/docs/_spec/openapi.yaml, linked as 'OpenAPI Spec' from the API overview page (25). No llms.txt on kagi.com or help.kagi.com, both 404. The docs' Markdown source is in a public GitHub repository, which is not the same as serving it to agents (0). Parameter descriptions say what each does and which option wins when lens, filters and query operators overlap, and the MCP tool says when to pick each workflow (14). Enums for workflow, format, time_relative and rule kinds, ranges on page, limit, timeout and extract.count, and one required field. Result props and meta are open objects (13). Three request examples with full responses and one 401 example. The error envelope is typed, but there is no list of error codes, and the file says its error documentation URLs return 404 for now (9). The path carries /v1 and v0 is documented as legacy. The changelog is product-wide, with API items on 21 May, 16 June and 21 August 2026 and none in the 10 September or 16 October posts (10). 25 + 0 + 14 + 13 + 9 + 10 = 71.

Score by category

CategoryWeightHyperresearchKagi Search APIEdge
Reliability16%3543Kagi Search API +8
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%7671Hyperresearch +5
Agent ergonomics13%8375Hyperresearch +8
Security & auth14%7964Hyperresearch +15
Payments & pricing10%6030Hyperresearch +30
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8841Hyperresearch +47
Transparency & trust7%7174Kagi Search API +3
Negative events≤1500
TotalB 68.1/100C 57.3/100

Facts side by side

FactHyperresearchKagi Search API
KindHTTP APIHTTP API
VendorHyperresearchKagi
Hosted endpointhttps://answers.hyperresearch.ai/mcphttps://kagi.com/api/v1/search
TransportsStreamable HTTPHTTP, Streamable HTTP, stdio
AuthOAuth or keyAPI key
PricingPay per usePay per use
Price for web search$0.05 per call$12 per 1,000 requests
x402nono
LicencenoneProprietary service under the Kagi API Terms of Service. The MCP server is MIT and the OpenAPI file states Apache-2.0
Read-only variant documentednono
llms.txtyesno
MCP registryai.hyperresearch/answersnot listed
Last release2026-11-162026-08-21
Terms last updated2026-11-172026-09-22
Privacy policy last updated2026-11-17no date given
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textyes
Terms restrict benchmarkingyesyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textnot found in the text
Popularitynone535 stars

How the scores have moved

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.

Kagi Search API

A public OpenAPI file, typed filters and API keys that can be limited by product and IP address make the API simple to call, at $12 per 1,000 searches. It is still a public preview, with no published rate limits and no free allowance, and the terms forbid storing results. Its status page logged only regional incidents since August.

Before you call either

Hyperresearch

  1. Use Answers (data_and_web_answer with mode auto) for one fact, and start a Light or Deep run only when the user wants a written report
  2. Send an Idempotency-Key on every write and reuse it only with an unchanged body, so a retried run or answer is charged once
  3. 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
  4. If a response has retryable: false, change the request rather than retrying. Branch on error.code, never on the message
  5. Use a key that starts hr_test_ to build and test an integration without spending

Kagi Search API

  1. Send POST https://kagi.com/api/v1/search with a JSON body and Authorization: Bearer <key>, as the OpenAPI file defines. Only query is required
  2. Set limit to cap results returned. In the mixed search workflow the MCP tool applies the cap to each result category separately
  3. Use extract.count (1 to 10) only when page text is needed. Extraction is billed at the Extract rate on top of the search
  4. Do not store results. The API terms allow only the transient caching an application needs, and require deletion on termination
  5. On 429 back off before retrying. The response covers both rate limiting and an exhausted usage limit, and no Retry-After header is documented. Check status.kagi.com for regional incidents

Questions

Which is better for AI agents, Hyperresearch or Kagi Search API?

Hyperresearch scores 68.1 (B) on agent readiness against Kagi Search API's 57.3 (C), and leads in 5 of 7 scored categories. Kagi Search API leads on reliability.

Do Hyperresearch and Kagi Search API need an API key?

Hyperresearch takes an API key or an OAuth sign-in. Kagi Search API needs an API key.

Can an agent call Hyperresearch and Kagi Search API without installing anything?

Yes. Hyperresearch has a hosted endpoint at https://answers.hyperresearch.ai/mcp and Kagi Search API at https://kagi.com/api/v1/search.

Other comparisons with Hyperresearch or Kagi Search API

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.