How we rate

Every score on this site is about one question: how well does this tool work when an AI coding agent builds with it? Here is exactly how we answer it.

Per-agent scores (1–5)

Each tool gets a separate score for each agent: Claude Code, Codex, Cursor, GitHub Copilot and Antigravity. We look at whether the vendor ships an official MCP server, whether it supports remote connections with OAuth, whether there is an official plugin or skill, and whether the vendor documents setup for that agent.

Evidence or nothing

A score is published only with at least one source link (vendor docs, a marketplace listing or the plugin's files) and the date we checked it. If we have not verified a tool in an agent yet, the table says "Not yet verified". We never estimate or fill gaps.

Agent-Ready score

A usage-weighted average of a tool's verified per-agent scores, weighted by how many developers use each agent (Stack Overflow Developer Survey 2026). It only appears once at least two agents are verified.

Trust signal

Whether the integration is official (built or listed by the vendor or the agent maker) or community-built. Unofficial MCP servers can be fine, but you should know which one you're installing.

MCP token cost

How many tokens a server's tool definitions take up in your context before you type anything. We publish it only once measured, with the date.

Freshness

Agents change every month. We re-check every agent monthly and show the last verified date on each agent page and score.

Thin pages stay hidden

Pages without enough verified data are marked noindex and left out of the sitemap until they are useful.

Money never buys a score

We earn from labeled sponsorships, paid fast-track reviews and some affiliate links (marked rel="sponsored"). None of these change a score or a ranking. A paid review gets reviewed sooner, not rated higher.

Spotted something wrong? Send a correction.