MCP servers for web fetching, search engines, crawling and content extraction. Each entry is scanned with the deterministic Capability-Flow Trust Model — grades are computed, never self-reported.
Modular, extensible integration with DataForSEO API endpoints, supporting both HTTP and SSE transports.
A fetch API based, remote server first, TypeScript SDK for MCP.
Shared scaffolding for the chrischall MCP fleet — server bootstrap, tool-result formatting, helpful errors, hardened env/config, a bearer API-client kit, zod atoms, session registries, a fetchproxy transport adapter, auth resolver skeletons, an in-memory
MCP Server for Brazilian ICD-10 (DATASUS) - International Classification of Diseases
Model Context Protocol server for Rollbar
The official Google Maps server: geocoding, places and directions data for assistants.
MCP server for Mealie recipe management
MCP server for Gong.io - access calls, transcripts, and users
Shared MCP server building blocks for @us-all servers — token-efficient by design (category toggles, extractFields, search-tools meta, registry, runtime)
A Model Context Protocol (MCP) server for Reddit with full read AND write operations - create posts, reply, edit, and delete content.
Fetch an entire site and use it as an MCP Server
A Model Context Protocol server with Desearch for real-time AI search, X search and web search.
Unofficial pi package that exposes Z.ai MCP server tools for web search, URL reading, repository reading, and vision workflows.
Returns token-optimized, structured YouTube data using the YouTube Data API v3.
Xquik agent skill & plugin bundle for REST, MCP, webhooks, exports & confirmation-gated X workflows. Not affiliated with X Corp.
A Model Context Protocol (MCP) server that provides search and crawl functionality using Search1API
Searches the Metropolitan Museum of Art collection and exposes artworks as resources.
Suprsonic MCP server. Gives any AI agent dozens of capabilities through one connection.
MCP server for Forge — Voxell's text-embedding API (turbo→ultra; ultra = Qwen3-Embedding-8B, ~75+ avg MTEB). embed + list_models for semantic search & RAG.
ESPN MCP — keyless multi-sport live scores, teams, and news via ESPN's public site API.
Community-built MCP server for Google Search Console API
MCP server for The Mountaineers website (mountaineers.org)
MCP server for the UK Companies House API — connects Claude, Cursor, Zed, and other AI tools to live company data
MCP server for the Open Library
A search index queried mid-conversation, a page fetched and stripped to clean text, a headless browser rendering JavaScript that no plain HTTP request could reach — that is the territory this category covers. These MCP servers use the Model Context Protocol to expose search engines and APIs like Brave Search, Tavily, Exa and SearXNG, simple fetch utilities, browser-automation stacks such as Playwright and Puppeteer, and Firecrawl-style crawling frameworks as tools that Claude, Cursor and other LLM clients can call directly. The result is an assistant that answers from the live web instead of stale training data. Browse the full MCP Trust Registry to compare candidates side by side.
Every fetched page is untrusted input, which makes this the category most exposed to prompt injection: a poisoned page can smuggle hidden instructions into the model's context, right next to yours. An over-permissioned fetch tool also doubles as a server-side request forgery hazard, reaching localhost services, internal dashboards or cloud metadata endpoints unless URLs are restricted. Attacker-shaped links can quietly exfiltrate conversation data through query strings, and headless-browser servers pile form submission and arbitrary script execution on top. Each listing therefore carries a Trust Score from the deterministic scoring engine, which reads the published package, maps its capability blast radius and toxic-flow paths, and grades it A–F — identical input always yields the identical grade, and no ranking can be bought.
Look for tight network scoping first: domain allowlists, no write-capable tools bundled beside fetch, and a strong Trust Score backed by clean security findings. Treat everything a server retrieves as data, never as instructions to follow. Reviewing the evidence attached to each registry listing before installing takes minutes and rules out the worst offenders.