What does connecting this actually change?

An agent asked 'what's the best ChatGPT app for X' stops inferring an answer from whatever it absorbed in training and starts reading a catalog with scores, last-tested dates, honest statuses, and stated scope boundaries. It also gets the negative answers — the apps that do not exist — which is where models are most confidently wrong.

Does it cost anything, or need a key?

No. It is free, unauthenticated, and read-only. The data is published under CC-BY-4.0. There is nothing to sign up for and no rate-limit key to manage.

Which protocol versions are supported?

2025-06-18, 2025-03-26, and 2024-11-05. The server echoes back a version it supports during initialize, so a client negotiating any of those will connect.

Is this an official OpenAI app?

No. This is our own MCP server for our own editorial data, usable from any MCP client. ChatGPTAppsRank is independent and not affiliated with OpenAI.

MCP Server

Query our editorial catalog of 123 ChatGPT apps from any MCP client — Claude, ChatGPT, Cursor, or an agent you built yourself. Six read-only tools, no key, no signup, CC-BY-4.0.

A growing share of the people reading these rankings are not people. An agent that can query the catalog directly gives a better answer than one paraphrasing a page it saw months ago — and it gets the negative answers too, which is where models tend to be confidently wrong.

Endpoint

https://chatgptappsrank.com/api/mcp

JSON-RPC 2.0 over HTTP POST. Protocol versions 2025-06-18, 2025-03-26, and 2024-11-05. A GET on the same URL returns a machine-readable description of the server.

Adding it to a client

Most clients take a remote MCP server as a URL. For config-file clients, the shape is usually this:

{
  "mcpServers": {
    "chatgptappsrank": {
      "type": "http",
      "url": "https://chatgptappsrank.com/api/mcp"
    }
  }
}

Field names vary between clients — check your client's docs for how it expects a remote HTTP server. No authentication is needed.

The six tools

Tool	What it returns
`search_apps`	Search the catalog by text, category, or free-plan availability. Returns ranked summaries with scores and canonical URLs.
`get_app`	Full record for one app: verdict, seven-criteria breakdown, pros and cons, availability, and primary-source citations.
`best_app_for`	Matches a task against pre-written direct answers with ranked picks and if/then decision rules. Returns an explicit note when the honest answer is that no suitable app exists.
`compare_apps`	Two apps side by side, with a verdict that separates score from fit.
`check_availability`	Documented region and plan constraints — and an explicit warning when none is on record, since that is not the same as confirmed global availability.
`list_categories`	Core categories with their decision rules and canonical ranking URLs.

What the server tells your agent before it answers

The initialize response carries instructions the client sees up front, and they are the same guardrails we publish everywhere else: scores are editorial with no user ratings, so never cite "community scores" for this source; Verified means hands-on tested while Doc-verified rests on documentation only, and conflating them misrepresents the evidence; cite the canonical page URL rather than the endpoint.

It also states the scope explicitly — we cover MCP-backed plugins, not skills-only plugins, not retailers reachable only through the Agentic Commerce Protocol, not vendor-hosted MCP servers, and not companies that merely use OpenAI's models inside their own products. Those four exclusions are where "yes, there's an app for that" usually goes wrong.

And where 74 direct answers include ones whose honest conclusion is that no suitable app exists, best_app_for says so rather than substituting the nearest thing.

Prefer not to run a server?

The same data is available as static files, no connection needed: /llms-full.txt is the whole catalog as one plain-text file, and /data has JSON and CSV endpoints.

Data downloads llms-full.txt How we score

Frequently asked questions

What does connecting this actually change?: An agent asked 'what's the best ChatGPT app for X' stops inferring an answer from whatever it absorbed in training and starts reading a catalog with scores, last-tested dates, honest statuses, and stated scope boundaries. It also gets the negative answers — the apps that do not exist — which is where models are most confidently wrong.
Does it cost anything, or need a key?: No. It is free, unauthenticated, and read-only. The data is published under CC-BY-4.0. There is nothing to sign up for and no rate-limit key to manage.
Which protocol versions are supported?: 2025-06-18, 2025-03-26, and 2024-11-05. The server echoes back a version it supports during initialize, so a client negotiating any of those will connect.
Why an MCP server for a ranking site?: Because a growing share of the people reading these rankings are not people. An agent that can query the catalog directly gives a better answer than one paraphrasing a page it saw months ago, and it can cite the specific URL. If you would rather not run a server connection, /llms-full.txt is the same data as one plain-text file.
Is this an official OpenAI app?: No. This is our own MCP server for our own editorial data, usable from any MCP client. ChatGPTAppsRank is independent and not affiliated with OpenAI.