Skip to content
Siftwright logo

Siftwright MCP server

Which servers in the official MCP registry are Siftwright's own, whether they answer, and what they offer. A program checked each of these, nobody copied them from a product page.

read 2026-10-25

← Back to Siftwright

Whose server it is
  • Registry namecom.siftwright/web-dataverified in the MCP registry

The registry name sits under siftwright.com. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

When we asked itchecked 2026-10-25
  • Stateanswerschecked 2026-10-25
  • Tools9
  • Server calls itselfsiftwright
  • Server version1.0.0
  • Protocol version2025-06-18

On 2026-10-25 the server answered initialize and tools/list without a sign-in. First answer in 17 ms.

Should an agent connect itNot proven either way yet
  • Its tools do every use case we checked: Fetch YouTube transcripts as plain text or timestamped segments; Capture full-page or mobile screenshots of public URLs; Pull Google News articles for a keyword or topic.
  • Checked on 2 of the 7 days it takes to judge reliability.

Once connected

  • For "Fetch YouTube transcripts as plain text or timestamped segments", the likely first call is youtube_transcript.
  • For "Capture full-page or mobile screenshots of public URLs", the likely first call is website_screenshot.
  • For "Pull Google News articles for a keyword or topic", the likely first call is google_news.
  • Parameters are untyped or undescribed in 3 of 9 tools: check arguments before calling.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

Measured dailydaily checks: 2, the first on 2026-10-25
  • INITE MCP scoreafter a weeka score needs 7 daily checks: 2 so far
  • What an agent can do with it3 of 3 use cases
  • Answeredmeasuring2 of the 7 daily checks it takes to print a share
  • Median first answer148 ms
  • Space in an agent's context≈ 1,805 tokensthe definitions of all its tools (9), which an agent carries on every turn
  • Tool definitionsA · 94 of 100
  • Typed results0 of 9 toolsdeclared outputSchema: results come typed, read once
  • Parameters without a description: in 3 of 9

What an agent can do with it

  • can do · Fetch YouTube transcripts as plain text or timestamped segments
  • can do · Capture full-page or mobile screenshots of public URLs
  • can do · Pull Google News articles for a keyword or topic

Each of the product's use cases, put to Jev against this server's tool list: can an agent do it with these tools?

Worth a look

None of the five warning signs in these definitions.

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (30%), doing the product's job (25%), definitions (20%), context (10%) and use (15%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

Tools it offers9 tools · read 2026-10-25
  • youtube_transcript

    Get transcripts (captions) for YouTube videos, Shorts, playlists or channels as plain text and timed segments, with video metadata. Uses the captions YouTube already has; videos without captions return a failed item. Each video whose…

  • website_screenshot

    Capture a PNG or JPEG screenshot, or render a PDF, of public web pages in a real browser. Returns a signed fileUrl per page that can be downloaded without a key for 7 days. Each page captured counts as one result.

  • google_news

    Search Google News for a keyword, company or topic and get articles as structured data (title, link, source, published time). Each article returned counts as one result.

  • pqc_scan

    Check whether domains use post-quantum TLS key exchange (X25519MLKEM768 / ML-KEM), which groups they prefer, TLS 1.2/1.3 support, certificate algorithms and expiry, and HSTS, with a letter grade. Each host that answers counts as one result.

  • contact_details

    Find public emails, phone numbers and social profiles (LinkedIn, X/Twitter, Facebook, Instagram, YouTube) on company websites by crawling up to a few pages of each site. Each website with a successful crawl counts as one result.

  • tech_stack

    Detect what a website is built with: CMS, ecommerce platform, frameworks, analytics, ad pixels, hosting, CDN, email and DNS provider, with the evidence for each detection. Each website that loads counts as one result.

  • app_store_reviews

    Get Apple App Store reviews for apps (by name, numeric ID or URL) in one or more countries, filtered by rating, keywords and date. Every 25 reviews returned count as one result (rounded up per request).

  • ats_jobs

    List open jobs from company careers pages on Greenhouse, Lever, Ashby, SmartRecruiters and Recruitee in one schema (title, team, location, remote, salary when published, apply URL), with title, location, department and recency filters…

  • domain_whois

    Look up who a domain is registered with, when it was created, when it expires, its status, nameservers and DNSSEC (from RDAP/WHOIS registries), plus A, MX, SPF and DMARC records, and whether it looks available. Every 3 domains with a…

How to connect

Add the server to your MCP client config. Claude Desktop, Cursor and most other clients read this shape:

{
  "mcpServers": {
    "web-data": {
      "type": "http",
      "url": "https://siftwright.com/mcp"
    }
  }
}

Add it

Claude Code, in a terminal

claude mcp add --transport http siftwright https://siftwright.com/mcp

Add to CursorAdd to VS Code

claude.ai, ChatGPT and other clients: add https://siftwright.com/mcp as a custom connector in the settings.

OpenAPI spec

9 paths, read 2026-10-25. Open the spec ↗

Each line is something a program found when it went looking: the registry entry, the server's own answer, a spec that parses. A missing line means we did not find it, not that it does not exist.