RightAIChoice — verified AI tool data

What changed, what died, what it costs, what survives: 8,000+ AI tools, verified and dated.

Hosted MCP Server

npx add-mcp 'https://rightaichoice.com/api/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Free · no API key · read-only

The MCP server for AI tools: verified status, pricing and sentiment for any AI assistant

Connect Claude, ChatGPT, Cursor or any MCP-capable agent to the RightAIChoice verification engine and it stops recommending dead AI tools. Thirteen read-only tools answer the questions that matter before adopting anything — is it alive, what does it really cost, what do users think, what should I use instead — and every answer is dated: when we last probed the vendor's links, and when we last rebuilt our record.

Endpoint

https://rightaichoice.com/api/mcp

Add to Claude Code

claude mcp add --transport http rightaichoice https://rightaichoice.com/api/mcp

README & citation · Listed on the official MCP registry as com.rightaichoice/mcp

AI tools under verification

8,123

Dead homepages right now

169

PH launches dead (of known outcomes)

24.4%

Re-verification cycle

Weekly

Live figures as of 2026-11-01; survival study probed 2026-07-18.

Why AI assistants need a verification layer

LLMs recommend products that don't exist. Slopsquatting research measured 19.7% of packages recommended by code models as nonexistent — over 205,000 phantom names.
The AI tool market has real mortality. Our survival study of 2,291 top Product Hunt launches (2023–2025) found that of the 2,066 with a determinable outcome, 24.4% are already dead. Training data never catches up: a model's knowledge of "good AI tools" is months to years stale.
One wrong answer ends trust. An assistant that recommends a shut-down product, quotes retired pricing, or invents a tool loses the user. One MCP call before the recommendation prevents all three.

The thirteen tools

check_tool_status

"Is this AI tool still alive?"

Healthy / monitor / at-risk verdict from weekly probes of every vendor link, with the last-verified date and evidence.

compare_tools

"X vs Y — which should I pick?"

Live side-by-side verification of both tools plus our reviewed editorial verdict when one exists for the pair.

find_alternatives

"What should I use instead?"

Same-category alternatives, each carrying its own health verdict, verified age and pricing — never unverified filler.

market_sentiment

"What do users actually think?"

Synthesized real-user reports from Reddit, HN, YouTube, Product Hunt and more — per-source positivity, top pros and cons, scan date.

pricing_reality

"What does it really cost?"

Plan-by-plan prices exactly as the vendor lists them, plus documented hidden costs: annual markups, seat minimums, paywalled features.

viability_score

"Is it safe to build on?"

A viability band (safe bet / moderate / at risk) with five measured drivers and the weakest signal explained in plain language.

deadpool_digest

"How many AI tools die?"

Live catalog decay rates by category plus our frozen survival study of 2,291 Product Hunt launches. Every figure dated.

recommend_tools

"Find me a tool that does X — that won't disappear"

A verified shortlist matched on what you need it to DO, screened against a survival bar so nothing on the list is already failing our checks. Tools with no measurement yet are named, never silently dropped.

category_landscape

"What does this whole category look like?"

How crowded a category is, how its tools distribute across survival bands, its vendor-link decay rate, and named examples at both the strongest and weakest ends.

stack_overlap

"Are we paying twice for the same thing?"

Which pairs in your stack we classify the same way, with the shared categories named so you can judge it yourself. A prompt to review a pair — never a recommendation to cancel one.

check_mcp_server

"Is this MCP server listed, and does it respond?"

The official registry record for an MCP server — status, version, how stale the listing is, declared endpoints — plus one anonymous request to see whether it answers. Observations only; never a safety or trust assessment.

whats_changed

"What changed in the AI tool market?"

Dated, field-level changes we actually observed — vendor pricing-page edits, price changes, link health, domains and certificates. Our own editorial rewrites are excluded on purpose.

best_for

"What are the best AI tools for X?"

Our ranked list for a topic — topic match, real-user mentions, AI-engine reads and viability, gated by a reviewed consensus list. A top pick is named only when the leader clears that gate.

What an answer actually looks like

Not a mock-up. The block below was produced by the live server when this page was last rebuilt, through the same code path an assistant calls — so what you read here and what your assistant receives cannot drift apart.

Request

check_tool_status { "tool": "Jasper" }

Response

Jasper — verdict: HEALTHY
Jasper is alive and actively maintained — main site and official links responding. Verified Oct 30, 2026.
Freshness — two clocks, not merged: vendor links last probed today (rolling weekly probe cycle) — this is what the alive/at-risk verdict is built on; catalog data (pricing, features, verdict prose) last rebuilt 3 days ago. Method: automated link/liveness probes of every vendor URL + a 5-signal viability model.
Source: https://rightaichoice.com/tools/jasper?utm_source=mcp&utm_medium=llm&utm_campaign=check_tool_status

Note the two dates and the method behind each. That pairing is the whole point: a verdict you can date is worth something, an undated one is not.

Connect in under a minute

1. Claude Code (one command) claude mcp add --transport http rightaichoice https://rightaichoice.com/api/mcp
2. Claude Desktop / claude.ai Customize → Connectors → + Add → Add custom connector → paste https://rightaichoice.com/api/mcp
3. Cursor (mcp.json) { "mcpServers": { "rightaichoice": { "url": "https://rightaichoice.com/api/mcp" } } }
4. ChatGPT chatgpt.com/plugins → + → Add custom MCP server → https://rightaichoice.com/api/mcp
5. Windsurf (mcp_config.json) { "mcpServers": { "rightaichoice": { "serverUrl": "https://rightaichoice.com/api/mcp" } } }
6. VS Code (GitHub Copilot agent mode) .vscode/mcp.json: { "servers": { "rightaichoice": { "type": "http", "url": "https://rightaichoice.com/api/mcp" } } }
7. Agent SDKs (OpenAI Agents, Vercel AI SDK, LangChain) Point any streamable-HTTP MCP client at https://rightaichoice.com/api/mcp — no auth, no session setup needed.
8. Check it responds, from anywhere curl https://rightaichoice.com/api/mcp # returns {"status":"ok", ...} — use this for uptime monitoring

Works with any MCP client over streamable HTTP — including the OpenAI Agents SDK, Vercel AI SDK and LangChain MCP adapters. No local install; it's a remote server.

The honesty contract

Every answer carries two dates, kept apart — when we last probed the vendor's links, and when we last rebuilt our record — plus the method behind each. Never an undated claim, and never one date standing in for the other.
Unknown tools are reported as not in the catalog, which is not evidence they are dead.
Fuzzy name matches are disclosed — the server names what it matched so your assistant can reject a wrong match.
"Vendor publishes no prices" and "we haven't recorded prices" are two different answers, never merged.
Viability comes as a band with named drivers, not a leaderboard number.
Live decay statistics and the dated survival study are never mixed in one figure.

The server reads from the same engine as the public tool checker and the AI Tools Deadpool report, so an MCP answer can never contradict the website. Method details: methodology.

Why connect this instead of trusting the model

An assistant without this already answers questions about AI tools. The question is whether the answer is true. Three things it cannot do on its own:

It cannot know what happened after training

A model’s picture of "good AI tools" is months to years old. Products shut down, change hands, triple their prices and quietly drop free tiers in that gap — and nothing in the model tells it which. Our records are re-probed on a rolling weekly cycle and every answer is dated, so the assistant can see the age of what it is repeating.

It cannot tell a real product from a plausible name

Research on "slopsquatting" measured 19.7% of packages recommended by code models as nonexistent — over 205,000 phantom names. A model generates a name that looks right. One call separates "this exists and is alive" from "this reads like it should exist".

It cannot say "I do not know" about a specific tool

This is the one that matters most. Asked about something it has no information on, a model fills the gap. Our server returns "not in the catalog — and that is NOT evidence it is dead", which is a different sentence from both "it is fine" and "it is gone", and the only honest one.

What to ask it, in your own words

You do not call these by name — you ask the assistant a question and it picks the tool. These are the questions each one is built to answer.

“Is this AI tool still alive?”

Healthy / monitor / at-risk verdict from weekly probes of every vendor link, with the last-verified date and evidence.

check_tool_status

“X vs Y — which should I pick?”

Live side-by-side verification of both tools plus our reviewed editorial verdict when one exists for the pair.

compare_tools

“What should I use instead?”

Same-category alternatives, each carrying its own health verdict, verified age and pricing — never unverified filler.

find_alternatives

“What do users actually think?”

Synthesized real-user reports from Reddit, HN, YouTube, Product Hunt and more — per-source positivity, top pros and cons, scan date.

market_sentiment

“What does it really cost?”

Plan-by-plan prices exactly as the vendor lists them, plus documented hidden costs: annual markups, seat minimums, paywalled features.

pricing_reality

“Is it safe to build on?”

A viability band (safe bet / moderate / at risk) with five measured drivers and the weakest signal explained in plain language.

viability_score

“How many AI tools die?”

Live catalog decay rates by category plus our frozen survival study of 2,291 Product Hunt launches. Every figure dated.

deadpool_digest

“Find me a tool that does X — that won't disappear”

recommend_tools

“What does this whole category look like?”

How crowded a category is, how its tools distribute across survival bands, its vendor-link decay rate, and named examples at both the strongest and weakest ends.

category_landscape

“Are we paying twice for the same thing?”

Which pairs in your stack we classify the same way, with the shared categories named so you can judge it yourself. A prompt to review a pair — never a recommendation to cancel one.

stack_overlap

“Is this MCP server listed, and does it respond?”

check_mcp_server

“What changed in the AI tool market?”

Dated, field-level changes we actually observed — vendor pricing-page edits, price changes, link health, domains and certificates. Our own editorial rewrites are excluded on purpose.

whats_changed

“What are the best AI tools for X?”

Our ranked list for a topic — topic match, real-user mentions, AI-engine reads and viability, gated by a reviewed consensus list. A top pick is named only when the leader clears that gate.

best_for

Every one is read-only and annotated as such, so assistants like Claude call them without asking you to approve each request.

What this server refuses to do

Every catalog will tell you what it returns. These are the answers we deliberately will not give, because the data behind them does not support the confidence the answer would imply. Each one is enforced in code and covered by a test.

No overall score for a stack. Ask about twenty tools and you get a count and a named list, never an average. Nine healthy tools and one corpse do not average to "mostly fine" — the corpse is the finding, and a single number hides exactly the thing you needed.
No currency conversion, ever. Prices are returned exactly as the vendor published them, in whatever currency and cadence they used. A wrong conversion inside a buying recommendation is real financial harm, so there is no "under $X" filter at all.
No "cancel this tool". The overlap tool shows which products we classify the same way and names the shared categories so you can judge it. It does not know your seats, contracts or which team depends on what, so it recommends nothing.
No silent omissions. Tools excluded for having no measurement yet are listed by name, with the reason. Dropping them quietly would be a hidden editorial act dressed up as a filter.
No security or trust verdict on other software. We can observe whether something responds. We cannot observe what it does with your data, so we never publish a verdict that implies we checked.
Our own edits are not market news. When we rewrite a description, that is us changing — not the market. Those changes are excluded from "what changed", and the response says so.

Coverage, and where it ends

The limits matter as much as the coverage. An assistant that knows where our data stops can tell you so, instead of guessing past it.

Matching understands meaning — and says when it could not

When you describe a need rather than name a product, we combine meaning-based search (text embeddings of every listing) with keyword search, so different wording for the same job still finds it. Every answer states which methods actually ran; if meaning-based search was unavailable, the answer says it was keyword-only.

The change record has a start date

Field-level change history begins on a fixed date, stated in every answer. Ask about an earlier period and you get that date back, not an empty result dressed up as "nothing changed".

Not in the catalog is not dead

A tool we do not hold may be new, niche, or not an AI tool. Every miss says so explicitly, because "we have no record" and "it shut down" are completely different facts.

Categories are ours, and they overlap

Our taxonomy is a classifier’s output, not an objective fact, and a tool sits in several categories at once. Category sizes therefore overlap and never sum to the catalog total.

Bands classify risk, not quality

A survival score measures the chance a product disappears. It has no notion of company size, so a small well-run tool can outscore a household name. It is not a leaderboard and must not be read as one.

Read-only, and nothing is live-scanned

No call an assistant makes can trigger a scrape, an AI call, or any change to the catalog. Answers come from data verified before you asked — which is why every one carries a date.

Frequently asked questions

What is the RightAIChoice MCP server?

A free Model Context Protocol (MCP) server at https://rightaichoice.com/api/mcp that gives any AI assistant — Claude, ChatGPT, Cursor and every MCP-capable agent — verified answers about 8,000+ AI tools: live status, pricing reality, user sentiment, viability and alternatives. Every vendor link is re-probed on a rolling weekly cycle (measured 2026-09-14: 100% of published tools probed within 7 days, median 2), and every answer states both that probe date and the separate, slower date our catalog record was last rebuilt.

Is it free, and do I need an API key?

Yes, free — and no API key or sign-up. The server is read-only and rate-limited per IP (120 requests/minute). All thirteen tools are annotated read-only, so assistants like Claude can call them without confirmation prompts.

Why should my AI assistant verify tools before recommending them?

Because models hallucinate products: research on "slopsquatting" found 19.7% of packages recommended by code LLMs do not exist, and our own survival study found 24.4% of 2,066 top Product Hunt launches from 2023-2025 are already dead. A one-call check prevents recommending a dead or imaginary tool.

How fresh is the data?

Two clocks, and we keep them apart. Every published vendor link (website, docs, changelog, repository) is re-probed on a rolling weekly cycle — that probe is what the alive/at-risk verdict is built on, and as of 14 Sep 2026 it had covered 100% of published tools within 7 days (median 2). Separately, the full catalog record — pricing, features, verdict prose — is rebuilt on a slower cycle, median 16 days. Mortality statistics recompute daily. Every MCP answer states both dates rather than merging them into one "last verified", so the assistant — and you — can judge each claim against the check that actually backs it.

What happens when a tool is not in the catalog?

The server says so honestly: "not in the catalog" with an explicit note that this is NOT evidence the tool is dead. When it resolves a fuzzy name it names the match it made, so the assistant can reject a wrong match instead of confidently answering about the wrong product.

Can it run live scans or modify anything?

No. All thirteen tools are read-only and serve pre-verified data from the RightAIChoice verification engine — the same engine behind rightaichoice.com. Nothing an assistant does through this server can trigger a scrape, an LLM call, or any change to the catalog. We do log the call itself — which tool, the query, the answer served, latency, and a salted hash of the client IP (never the raw IP) — for rate limiting and to see which tools people actually ask about.

Can I monitor whether it is up?

Yes. A plain GET to https://rightaichoice.com/api/mcp returns a small JSON health descriptor — status, the protocol version and the live tool count — so any uptime monitor can poll it directly. The MCP protocol itself is served over POST to the same URL, and a GET carrying "Accept: text/event-stream" is treated as a protocol request, so monitoring cannot interfere with clients. Health probes are not counted as usage.

Can it tell me if we are paying for two tools that do the same thing?

It can show you which pairs in your stack we classify the same way, naming the shared categories so you can judge the claim yourself. It is careful about what that means: it compares our classification, not the products’ actual features or what your team does with them. It never recommends cancelling anything and never totals cost, because plan and seat count are unknown to us.

What does "what changed" actually count?

Only changes the vendor made that we observed: pricing-page edits, recorded price changes, link health moving, domain and certificate changes, vendors moving their site, and site status flipping to unreachable, parked or redirected. Our own editorial rewrites are excluded on purpose — our writing changing is not the market moving — and the response says so. Our change record begins on a fixed date, stated in every answer.

Which AI assistants and clients support it?

Any MCP-capable client over streamable HTTP: Claude Code, Claude Desktop and claude.ai custom connectors, Cursor, Windsurf, ChatGPT (custom MCP servers), VS Code with GitHub Copilot, Cline, and agent SDKs like the OpenAI Agents SDK and Vercel AI SDK. No local install is needed — it is a remote server.

How do I add RightAIChoice to ChatGPT?

Go to ChatGPT Plugins (chatgpt.com/plugins).
Select the plus button, then "Add custom MCP server".
Enter a name (for example RightAIChoice) and a short description.
Enter the MCP server URL: https://rightaichoice.com/api/mcp
Our server needs no sign-in or API key. Review the risk warning and select "I understand and want to continue".
Select "Create as a plugin", then review the tools ChatGPT discovered from the server.

OpenAI states that account and workspace policies apply to adding custom MCP servers; if the option is missing, see OpenAI’s docs or ask your workspace admin.

Steps from ChatGPT's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to Claude?

Go to Customize > Connectors.
Click "+ Add", then "Add custom connector".
Enter a name (for example RightAIChoice).
Enter the remote MCP server URL: https://rightaichoice.com/api/mcp — then click "Continue".
Under "Authentication", choose "No sign in" — our server needs no account or API key.
Click "Add".

On the Free plan you can add one custom connector. On Team and Enterprise, an Owner adds it under Organization settings > Connectors, and members then find it under Customize > Connectors and click "Connect".

Steps from Claude's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to Claude Code?

Run the command below in your terminal.
To make it available in every project rather than only the current one, add --scope user. To share it with your team through the project’s.mcp.json, add --scope project.

Terminal

claude mcp add --transport http rightaichoice https://rightaichoice.com/api/mcp

Steps from Claude Code's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to Cursor?

Open (or create).cursor/mcp.json in your project, or ~/.cursor/mcp.json in your home folder to use it in every project.
Add the entry below and save.
Open Customize in the sidebar to check the server is enabled.

mcp.json

{
  "mcpServers": {
    "rightaichoice": {
      "url": "https://rightaichoice.com/api/mcp"
    }
  }
}

Steps from Cursor's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to VS Code and GitHub Copilot?

Open the Command Palette (Ctrl+Shift+P on Windows and Linux, ⇧⌘P on macOS) and run "MCP: Add Server".
Choose "HTTP (HTTP or Server-Sent Events)".
Enter the URL: https://rightaichoice.com/api/mcp and name it rightaichoice.
Or add the entry below to.vscode/mcp.json in your workspace (run "MCP: Open User Configuration" for the user-level file).

.vscode/mcp.json

{
  "servers": {
    "rightaichoice": {
      "type": "http",
      "url": "https://rightaichoice.com/api/mcp"
    }
  }
}

Steps from VS Code / GitHub Copilot's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to Windsurf?

Open Windsurf’s MCP configuration file, mcp_config.json — for where it lives in your version, see the client’s docs.
Add the entry below under mcpServers and save.

mcp_config.json

{
  "mcpServers": {
    "rightaichoice": {
      "serverUrl": "https://rightaichoice.com/api/mcp"
    }
  }
}

Steps from Windsurf's documentation, checked 1 Nov 2026.

How do I add RightAIChoice to Cline?

In the Cline panel, click the MCP Servers icon in the top toolbar.
Open the Remote Servers tab.
Enter a Server Name (rightaichoice) and the Server URL: https://rightaichoice.com/api/mcp
Choose the Streamable HTTP transport type and click "Add Server".
Or add the entry below to cline_mcp_settings.json — keep "type": "streamableHttp", because leaving it out falls back to the older SSE transport.

cline_mcp_settings.json

{
  "mcpServers": {
    "rightaichoice": {
      "type": "streamableHttp",
      "url": "https://rightaichoice.com/api/mcp"
    }
  }
}

Steps from Cline's documentation, checked 1 Nov 2026.