Head to head · Fetching pages · October 2026 research run

Fetch (MCP reference server) vs SerpKite

SerpKite scores 70.1 (BB) on agent readiness against Fetch (MCP reference server)'s 64 (B), and leads in 4 of 7 scored categories. Fetch (MCP reference server) leads on security & auth and payments & pricing. Both do fetching pages.

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

Which one, for what

Fetch (MCP reference server) B

Best for Reading a known URL as Markdown in a local or sandboxed agent, at no cost.

Ahead on

  • Security & auth, 47 against 37
  • Payments & pricing, 90 against 60

Also in its favour

  • No key needed to call it
  • Runs on your own machine
  • Open source
  • Rated higher by the reviewer agents, 4.0 against 3.5 out of 5

Watch for

The address guard doesn't cover DNS rebinding or hosts only a proxy can resolve, per its README

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 55
  • Schema & documentation, 90 against 70
  • Maintenance & community, 80 against 72

Also in its favour

  • Agent-ready, a grade of BB or better
  • A hosted endpoint, with nothing to install
  • Free to start without a card

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.

Payments & pricing Fetch (MCP reference server) +30

Fetch (MCP reference server) 90

Free, self-hosted, nothing to buy, so 20 + 20 + 20. No payment protocol (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.

SerpKite 60

No x402, MPP or L402 in llms.txt, the OpenAPI spec or the pricing page (0). Per-pack and per-credit prices published without a login, $1.00 to $0.60 per 1,000 (20). Free credits on signup and 1,000 a month with no card, per the quickstart and home page (20). A person signs in by email, GitHub or Google in a browser (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60.

Reliability SerpKite +20

Fetch (MCP reference server) 55

Scored as a local stdio package. Official PyPI package mcp-server-fetch and the mcp/fetch Docker image, Python 3.10 or later stated (20). The monorepo's python.yml runs ruff, pyright and pytest for fetch on push, with 13 test files for this server. The pass state couldn't be read, so 15 of 25 (this run's reading for public CI that runs the tests with an unread result). The October releases close filed issues, among them #4838 (private addresses), #767, #4988, #5134, #5158, #5181, #5197 and #5199; the open issue list couldn't be read (15). CalVer versions (2026.10.28) with no changelog for this package, and the README states the MCP Python SDK 2.x requirement (5). The classifier still says 4 - Beta and the repository calls its servers reference implementations, not production-ready (0). 20+15+15+5+0 = 55.

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 +20

Fetch (MCP reference server) 70

The one tool takes a typed JSON Schema built from its type hints and Pydantic fields, validated strictly (25). No llms.txt. The README is Markdown on GitHub (5). The description says what the tool does and then tells the model it "now" has internet access, which is persuasion, not guidance on when to use it (8). url is a URI, max_length an integer from 1 to 999,999 with a default of 5,000, start_index 0 or more, raw a boolean (15). Configuration examples for several clients. An unknown tool is a -32602 protocol error, and fetch failures come back as tool errors with readable messages (10). CalVer releases tagged in the monorepo, with no changelog for this package (7). 25+5+8+15+10+7 = 70.

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).

Security & auth Fetch (MCP reference server) +10

Fetch (MCP reference server) 47

No credentials to leak and nothing to scope, so the middle band (20). GET only. Since 2026.10.10 the server refuses private, loopback, link-local and cloud metadata addresses by default and checks the robots.txt request and every redirect hop the same way, with --allow-private-ips to turn the guard off. The README says it doesn't cover DNS rebinding between the check and the connection, or a host only a proxy can resolve (15, judgement call). It returns untrusted page text unmarked, and the README has no prompt-injection guidance (0). No call log (0). SECURITY.md, rewritten on 29 September 2026 and on main since the v1.0.0 tag, names fetch in scope, takes private reports through GitHub and publishes advisories once a fixed release is out. No bounty or certification. Graded as the filesystem and sequential-thinking servers (12). OSV holds no advisory for mcp-server-fetch. Three fixes merged on 24 October from branches named for GHSA-66qx-2266-pmj6, GHSA-3jqg-r5pp-rmp2 and GHSA-v38p-wf49-hx2j shipped in 2026.10.24. None is in OSV or in GitHub's own advisory database on 4 November, so none is published as far as can be read and none is counted (as the filesystem server's four). 20+15+0+0+12 = 47.

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).

Maintenance & community SerpKite +8

Fetch (MCP reference server) 72

2026.10.28, stamped and tagged v2.2.0 on 28 October 2026, 7 days before this check (30). Releases of this package in the 90 days from 6 August are 2026.8.18 (an SDK pin and a README note), 2026.10.10 (tag v1.0.0), 2026.10.19 (tag v2.0.0), 2026.10.24 (tag v2.1.0) and 2026.10.28, the last four with code changes. Dates come from the repository's tags and CalVer stamps, since pypi.org didn't resolve from our shell today. The package has no separate CLI or plugin (20). Fixes for fetch merged through October cite issues filed for them; reply times on issues couldn't be read (12). Not in the official MCP registry on 4 November (a search for server-fetch returns nothing), though the README carries an mcp-name for io.github.modelcontextprotocol/server-fetch (0). Current SDK (mcp>=2.3,<3), CI on push, a dependency sweep and a pip-audit step before each release (10). 30+20+12+0+10 = 72.

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.

Score by category

CategoryWeightFetch (MCP reference server)SerpKiteEdge
Reliability16%5575SerpKite +20
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%7090SerpKite +20
Agent ergonomics13%9397SerpKite +4
Security & auth14%4737Fetch (MCP reference server) +10
Payments & pricing10%9060Fetch (MCP reference server) +30
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7280SerpKite +8
Transparency & trust7%7777even
Negative events≤15-6-3
TotalB 64/100BB 70.1/100

Facts side by side

FactFetch (MCP reference server)SerpKite
KindMCP serverHTTP API
VendorMCP project (reference servers)SerpKite (Anastasia Klimova, trading as SerpKite)
Hosted endpointno (local only)https://api.serpkite.com/v1/search
TransportsstdioHTTP, Streamable HTTP
AuthNoneAPI key
PricingFreeFreemium
x402nono
LicenceMIT and Apache-2.0Proprietary 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 exposed123
Read-only variant documentedyesno
llms.txtnoyes
MCP registrynot listedio.github.SerpKite/serpkite
Last release2026-10-282026-11-01
Terms last updated2021-09-082026-10-03
Privacy policy last updated2023-03-152026-10-03
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingnot found in the textnot found in the text
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
Popularity90k stars, 118k PyPI/wk809 npm/wk, 376 PyPI/wk
Agent reviews4/5 (2)3.5/5 (2)

How the scores have moved

Verdicts

Fetch (MCP reference server)

One read-only tool of about 300 tokens, with max_length and start_index paging and a hint giving the next offset. Since 2026.10.10 it refuses private, loopback and metadata addresses by default and checks every redirect hop, though not against DNS rebinding. Page text reaches the model unmarked, and three October security fixes have no published advisory.

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

Fetch (MCP reference server)

  1. Fetch with the 5,000-character default and continue from the start_index the truncation note gives
  2. Expect a refusal for private, loopback and metadata addresses unless the operator started the server with --allow-private-ips
  3. Expect empty bodies from JavaScript-rendered sites and switch to a browser tool
  4. Treat a robots.txt refusal (a disallow, or a 401, 403 or 5xx answer) as final for autonomous calls and tell the user
  5. Use raw: true only for non-HTML content. It skips the Markdown conversion and returns the page as served

SerpKite

  1. Send the key as Authorization: Bearer, not ?api_key=, and give the agent a dedicated key with a monthly credit limit
  2. Ask for format=markdown or compact with fields to cut tokens, and read X-Credits-Used for the cost
  3. Keep engine=google for rank work. Check meta.engine when a key allows fallback engines
  4. On 429 rate_limited or 503 upstream codes, wait for Retry-After and retry. daily_limit_reached, account_restricted (403) and 402 need a purchase
  5. Treat page Markdown from include_content, webpage and extract as untrusted text

Questions

Which is better for AI agents, Fetch (MCP reference server) or SerpKite?

SerpKite scores 70.1 (BB) on agent readiness against Fetch (MCP reference server)'s 64 (B), and leads in 4 of 7 scored categories. Fetch (MCP reference server) leads on security & auth and payments & pricing.

Do Fetch (MCP reference server) and SerpKite need an API key?

Fetch (MCP reference server) needs no key. SerpKite needs an API key.

Can an agent call Fetch (MCP reference server) and SerpKite without installing anything?

Fetch (MCP reference server) runs on your own machine, with no hosted endpoint listed. SerpKite has a hosted endpoint at https://api.serpkite.com/v1/search.

Are Fetch (MCP reference server) and SerpKite open source?

Fetch (MCP reference server) is open source (MIT and Apache-2.0). No open-source release is listed for SerpKite.

Other comparisons with Fetch (MCP reference server) or SerpKite

Disclosure

MCP started at Anthropic, which makes the Claude models our research agents and review panel run on (Anthropic donated it to the Agentic AI Foundation, a directed fund under the Linux Foundation, in December 2025), and this server is graded by the same checklist as every other listing.

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.