"A task involving critical or creative thinking" matching MCP servers:

Enforces 21 critical rules derived from 96+ documented failure patterns to prevent common AI assistant mistakes. Provides automatic compliance checklists, rule summaries, and auto-updates.

MCP server enabling Claude Code to capture screenshots of Windows applications through WSL2, allowing AI to diagnose UI errors and verify layouts.

Enables AI agents to research App Store screenshots and human-reviewed references, then produce evidence-backed messaging, narrative, art-direction briefs, and measurable copy/export QA.

Enables Claude to independently check an AI agent's work in a local git repository by recording a baseline, running only the checks relevant to the changed files, and cross-checking what the agent claimed. Returns a compact, evidence-labelled report with a verdict and a ready-to-paste next prompt, so unverified success claims can't slip through.

A validation layer for AI coding assistants that enforces explicit LLM evaluations on plans, code diffs, and tests to ensure safer and higher-quality code.

Enables MCP clients to compile natural-language intent into a validated executable agent contract and to validate existing contracts through two exposed tools.

MCP-first A/B testing server that enables agents to manage experiments (create, update traffic splits, read results, apply winning variants) from tools like Claude Code, with self-hosted Cloudflare backend.

A multi-task MCP server for AI code optimization and testing that integrates with VSCode/Cursor, CLI tools, and autonomous agents. It provides tools for running tests, generating code improvements, and managing objectives across different environments.

Generates a pytest suite from acceptance criteria withheld from the coding agent, so the tests can disagree with the code; qikly_run never returns the criteria. The CLI also scores an existing suite by planting faults, and ships as an agent Skill (qikly --install-skill). Runs locally (uvx --from "qikly[mcp]" qikly-mcp): your code and API key stay on your machine.

Zero-dependency MCP server and CLI that proxies, inspects, and analyzes JSON-RPC message streams between MCP clients and servers.

A shared, concurrency-safe ledger for LLM code audits. It enables multiple agents to collaboratively report, verify, and resolve bugs with deduplication and hallucination resistance.

Enables agents to discover, search, compare, and inspect tested MCP tools by problem or keyword, with certification verdicts, scorecard summaries, certified scope, and explicit not-tested details.

Enables agents to inspect web pages for Cloudflare Turnstile challenges without an API key, and clear Turnstile or WAF challenges using a SolveGate API key.

Enables coding agents to verify whether a GitHub pull request is genuinely ready by checking draft state, conflicts, check results on the head commit, review requirements, merge queue status, and expected commit, returning a verdict with reasons and owners pinned to a commit and timestamp. It reports merged or closed PRs without judging them and never authorizes a merge.

Enables AI agents to hire real human operators for tasks requiring physical presence, human perception, or judgment, such as verification, testing, data collection, and physical-world tasks.

Measures what MCP servers cost in context tokens: probes a URL or a local command now, measures a whole config file, and looks up the daily results for every server in the official MCP registry on mcpcost.ulehla.dev.

Enables AI coding tools to run CheckMyLaunch scans on GitHub repos or live URLs and retrieve findings and repair prompts directly within the chat.

Lets an AI assistant statically scan HTML and JavaScript landing-page code that turns Meta's fbclid URL parameter into the _fbc cookie (and hand-written _fbp cookies), flagging known ways the click ID is changed, truncated, or kept from an older ad, without executing the code or contacting Meta.

Enables AI agents to generate Korea Government Design System components and page templates with KWCAG 2.2 accessibility requirements built in, then validate rendered pages and block task completion until remaining accessibility violations are corrected.

Lets AI agents ask which tests a Playwright or Vitest project contains, which are skipped and why, by reading the test source files statically. Nothing is executed, written, or sent over the network, and answers come back as paged JSON with summaries, filters, and scan diagnostics.

Enables AI agents to drive the 1C:Enterprise test client directly, opening forms, reading and modifying data, working with tables, and recording or replaying action scenarios. Connects to local or remote test clients over their network protocol without a testing manager.

Enables agents to measure a live page or Figma frame against a reference design, reporting exact numerical deviations in layout, spacing, typography, colors, states, and animations before any baseline exists.

MCP server that provides tools for evaluating LLM agent reliability, including adversarial task generation, automated LLM-as-judge assessment, and confidence statistics.

Email inboxes for AI agents: create a throwaway inbox, then block until the OTP or magic link arrives and get it back as a field. 7 tools, stdio, Python.

Translates a lockfile diff into a human-readable upgrade plan. For every dependency bump (npm or PyPI) returns semver class, breaking changes from GitHub release notes, CVEs fixed in the range, migration guide links, and a clear per-package recommendation. Bulk tool ranks up to 50 package changes in parallel by risk (security > caution > review > likely-safe > safe).

Enables AI agents to measure Chinese, Japanese and Korean page titles by rendered pixel width and judge whether they fit Google's search result column, flagging titles that would be truncated or cut near the edge. It also reads tags from local HTML files or directories and applies the same width, boilerplate and keyword-zone checks.

Enables MCP clients to run fresh audits of any public web page on mobile or desktop, returning category scores, lab metrics, and prioritized findings for performance, accessibility, best practices, and SEO. Each call performs a new uncached audit, so users can compare results within a conversation.

Enables an LLM or agent to drive a VUnit HDL unit-testing project end to end: list tests, compile, run, and inspect reports and per-test logs.

MCP server that exposes the JsonFabrica API as tools, so an AI agent can generate realistic synthetic JSON test data while it codes — single documents or full relational batches from reusable templates, reproducible with a seed. Runs locally over stdio; no network service.

Grades MCP servers on agent usability (A–F) — description quality, schema design, tool naming and token cost, the properties that decide whether a model picks the right tool. Benchmarked across 36 popular servers; a third scored D or F.