Catalog · EU AI Act checked · read-only MCP
Does this agent decide like your best person would?
Benchmarks score models on someone else's test. We measure one thing: how often an agent running a pack agrees with a named profession's expert on cases it has never seen, with the noise floor and the sample size printed next to the number.
agreement with a recruiter on 39 held-out cases, up from 40% without the pack
Simulated expert and rule-based agent on invented cases: it shows the instrument, not a real score. HeadhuntOS screening skill (synthetic demo) · full report
Catalog
Every pack, with its public facts and an AI Act checklist
Filter by status, profession or platform. Press / to search. Filters live in the URL, so a view can be shared.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): helps build agents with the Claude Agent SDK.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): maintains CLAUDE.md project instruction files.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): helps migrate code to current language and library versions.
First-party Claude Code plugin from the official marketplace that reviews a diff or PR for correctness bugs at a chosen effort level.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): simplifies and cleans up changed code without altering behaviour.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): commands for writing commits and preparing pushes.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): guided feature development from requirements to implementation.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): Go (gopls) language-server integration.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): creates Claude Code hooks from described automation rules.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): scaffolds and develops MCP servers.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): scaffolds and develops Claude Code plugins.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): a toolkit of pull-request review commands and agents.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): Python (Pyright) language-server integration for code navigation and diagnostics.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): Rust (rust-analyzer) language-server integration.
First-party Claude Code plugin that adds security guidance while coding; part of the official plugin marketplace.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository that applies a brand's visual and tone guidelines to generated material.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for designing visual canvases such as posters and one-page layouts.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository: reference for building on the Claude API and SDKs (models, tool use, streaming, caching).
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for co-authoring documents with a person, section by section.
Agent Skill from Anthropic's public skills repository for creating, reading and editing Word (.docx) documents, including tracked changes and comments. Source-available, not open source.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository with design guidance for building polished front-end interfaces.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for drafting internal communications such as announcements, updates and memos.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository that guides an agent through building an MCP server with tests and documentation.
Agent Skill from Anthropic's public skills repository for extracting, merging, splitting, filling and creating PDF files, including OCR of scanned pages. Source-available, not open source.
Agent Skill from Anthropic's public skills repository for creating and editing PowerPoint (.pptx) decks, templates, notes and comments. Source-available, not open source.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for creating, improving and evaluating other Agent Skills.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for generating consistent visual themes (colours, type) for generated interfaces and documents.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for building self-contained web artifacts (HTML, React) with a consistent structure.
Open-source Agent Skill (Apache-2.0) from Anthropic's skills repository for driving and testing a web application in a browser from an agent session.
Agent Skill from Anthropic's public skills repository for reading, editing and building spreadsheets (.xlsx, .csv) with formulas and formatting. Source-available, not open source.
First-party Claude Code plugin from the official plugin marketplace (anthropics/claude-plugins-official): TypeScript language-server integration for code navigation and diagnostics.
Apify Store actor that extracts public business listings from Google Maps; usable from agents through the Apify MCP server. Usage-based pricing on the Apify platform.
Cloudflare's product-specific MCP servers (Apache-2.0) for Workers, observability, DNS and other Cloudflare products, usable from any MCP client with a Cloudflare account.
Official Exa MCP server (MIT) for neural web search and crawling from an agent; usage-based pricing on the Exa API.
Official Firecrawl MCP server (MIT) giving agents web search, scraping and crawling tools; usage-based pricing on the Firecrawl platform.
Public, authless MCP server (Streamable HTTP) that tells a Polish employer which EU pay-transparency duties (Directive 2023/970) apply at its headcount, when the first pay-gap report is due, its risk score and the next fix. No employee data.
GitHub's official MCP server (MIT) giving agents tools over repositories, issues, pull requests and actions through the GitHub API.
Demo skill pack built with the expert-knowledge engine on synthetic cases only. Encodes a senior recruiter's confirmed screening rules; every verdict stays a human decision (HITL) and every outbound draft carries the AI Act Art. 50 disclosure.
Reference and test MCP server maintained in the official modelcontextprotocol/servers repository that exercises prompts, resources and tools; useful for testing MCP clients.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: fetches web content and converts it for the model.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: file operations with configurable access controls.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: tools to read, search and manipulate Git repositories.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: knowledge-graph based persistent memory for an agent.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: structured, revisable problem-solving through thought sequences.
Reference MCP server maintained in the official modelcontextprotocol/servers repository: current time and timezone conversion tools.
Official Playwright MCP server (Apache-2.0) that lets an agent drive a real browser through structured accessibility snapshots for testing and automation.
Official Notion MCP server (MIT) giving agents read and write access to Notion pages and databases through the Notion API.
Official Supabase MCP server (Apache-2.0) connecting AI assistants to Supabase projects: schema, SQL, migrations and project management.
Read-only MCP connector (OAuth) that shows where a workspace's LLM API money went by provider, model and feature, how much ties out to the invoice, and the exact config change for each fix. Aggregates only. Pre-launch: listed as submitted until the API is live.
MCP server (MIT) that serves up-to-date, version-specific library documentation to coding agents so generated code matches current APIs. Hosted service with free and paid tiers.
No pack matches. Submit one.
Method
Why we publish no score below 30 cards
Drag the slider. The green band is the 95% interval around an observed agreement rate. With ten cards the honest answer to "is this pack better than a coin with a bias?" is "we cannot tell", so the page says INDETERMINATE instead of printing a number.
The full method: golden set, noise floor, before and after, re-verification.
Wilson interval for a binomial proportion. Below 30 cards the interval is wider than the gap between a good pack and a coin flip with a bias; that is why we publish no number there.
Professions
Whose judgement is the yardstick
Reports, decks, spreadsheets and documents produced and checked by agents for business decisions.
Data collection, cleaning and reporting workflows run by agents and scrapers.
Cloud and LLM spend attribution, invoice reconciliation and cost fixes.
Employer-level duties under labour and pay-transparency law; deadlines, exposure and what to fix first. Not decisions about individual employees.
Screening, shortlisting and outreach decisions in recruitment. High-risk domain under the EU AI Act (Annex III, employment); a named human decides every outcome.
Code review, security review and frontend decisions encoded as agent skills and plugins.
For agents
Your agent can read this catalog without a browser
The same data as these pages, as JSON and as a remote MCP server. No key, no account, no tracking.
curl -s https://measuredagents.com/catalog.json | jq '.packs[] | select(.status=="verified")'curl -s -X POST https://measuredagents.com/mcp -H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_packs","arguments":{"query":"recruiter"}}}'Read-only, no key, no account. Tools: list_packs, get_pack, search_packs, list_professions. Add https://measuredagents.com/mcp as a remote MCP server in Claude, Cursor or any client that speaks Streamable HTTP.
List a pack, or ask for a measurement
Listing is free and takes a form. A verification run starts with a free 10-card sample from your profession's expert.