Confidence in the Truth
JERP fact-checks your AI before it answers: every claim comes back true, false, or not a fact, with a confidence score and a source. Your model reads less, so you pay for fewer tokens.
Works with your agent, not instead of it
How JERP checks your AI
What actually happens between your agent's question and its answer — in three steps.
- Token savings Your AI doesn’t read page after page to answer one question. JERP does the checking and hands back a short answer — your model reads less, you pay less.
- Strict answers Every answer is true, false, or not a fact — with how sure it is. No essay, no confident guess dressed up as research.
- No hallucinations If there’s no source for it, your AI can’t say it. Unsupported claims are cut before they reach you.
I’m launching a project-management tool for small agencies. Look at the top tools in this space and tell me which ones really have a free plan, which of them call themselves “the #1 choice for agencies”, and whether any of them actually integrates with Slack. I don’t want guesses — only what their own sites say.
- Top tools in this spacefound
- Has a free plantrue
- “#1 choice for agencies”not a fact
- Integrates with Slacktrue
Free plan and Slack: confirmed on their own sites. “#1 choice” is a slogan — not a fact, so it’s left out.
Connect it to your AI agent
It isn’t another chat window. Plug it into the agent you already use over MCP — pick yours below, copy the config, done — or call the REST API from your own code. Full steps and the tool list live in the docs.
After you click, the key expires in 1 hour
Then paste this into your agent
What changes in your AI’s answers
You keep asking in plain words. Your agent comes back with fewer, better answers — each one checked, sourced, and honest about what it doesn't know.
-
Real competitors Only sites that sell what you sell Your agent stops listing blogs, directories, and lookalike names as rivals. -
Checked claims No source — it isn't said Every fact in a draft is checked first. Anything without a source is cut. -
Fit to your market Does this site belong in your niche? A straight yes or no for the category you name, with the reason behind it. -
Strict answers A verdict, not an essay Every answer is true, false, or not a fact — and says how sure it is. -
Clean lists Real products in, noise out Paste any list of sites. Parked, dead, and placeholder pages are removed with a reason. -
Ranked for your question The most relevant site first Your agent orders sites by how well they answer what you asked — not by who shouts loudest. -
Your questions, per site Yes, no, or a score Ask the same thing about many sites — pricing, trials, audience — and get one clean answer each. -
Market map Everyone in your space, grouped Start from one site and see direct rivals, indirect ones, and platforms — each with evidence.
Pricing and token savings
Start free. Upgrade when your agent needs more room — same API, same tools.
-
Free $0/mo
Try JERP in chat, no card.
- 1-hour key, no signup
- 100 calls per key
- 1 call / second
- All 8 tools
-
Pro $5/mo
For a daily agent workflow.
- Persistent key, no 1-hour expiry
- Higher rate limit
- Watchlists and alerts
- Everything in Free
-
Enterprise $25/mo
For teams running it in production.
- Team seats, shared watchlists
- Highest rate limit
- Priority support
- Everything in Pro
Prices in USD. Pro and Enterprise are not billed yet — sign in free today and we'll email you when they open.
Tokens for one 24-site check
14.4× fewer tokens — JERP reads the pages, your model reads only the answer.
| Model | $ per 1M in / out | Without JERP | With JERP |
|---|---|---|---|
| AnthropicClaude Fable 5.1 | $10.00 / $50.00 | $0.73 | $0.11 |
| OpenAIGPT-6 Astra | $10.00 / $50.00 | $0.73 | $0.11 |
| AnthropicClaude Opus 5.5 | $4.00 / $20.00 | $0.29 | $0.046 |
| OpenAIGPT-5.6 Terra | $2.00 / $12.00 | $0.15 | $0.026 |
| GoogleGemini 3.1 Pro | $2.00 / $12.00 | $0.15 | $0.026 |
| xAIGrok 4.7 | $2.00 / $6.00 | $0.14 | $0.016 |
Questions about JERP
What JERP is — and what it is not — next to Ahrefs, Semrush, and AI visibility tools.
What is JERP?
JERP is a fact-checking layer between your AI and its answer. Your agent — Claude, ChatGPT, Cursor, or your own — asks JERP before it answers, and every answer comes back as true, false, or not a fact, with a confidence score and a source. It works over MCP and a REST API, and the free tier needs no account.
How does JERP reduce token costs?
Your model stops reading whole websites. JERP reads the pages on its side and hands back a short answer, so your model reads only the result. On one measured 24-site check that is about 4,700 tokens instead of 67,800 — roughly 14× fewer. The per-model cost table is in the Pricing section above.
How does JERP stop AI hallucinations?
No source, no claim. Every fact in an answer is checked against what the site itself says and against public records; anything JERP cannot source is cut instead of being guessed. When JERP is unsure, the confidence score says so rather than hiding it.
What do true, false, and not a fact mean?
True: a source confirms it. False: a source contradicts it. Not a fact: it is opinion or marketing — “the #1 choice”, “market leader” — and cannot be checked, so your AI should not state it as fact. Each verdict carries a confidence score.
Is JERP a replacement for Ahrefs or Semrush?
No. Those tools crawl the web for keywords, links, and rankings. JERP checks the sites and claims your agent is already working with — real competitor or noise, fits your niche or not, sourced or not. Many teams use both.
Does JERP measure brand mentions in ChatGPT or other AI answers?
No. JERP does not track how often a brand appears in AI answers. It makes the answers your own agent gives you checkable.
Do you return Google rankings?
No. JERP orders sites by how well they answer your question, not by search position. It is not a SERP tracker.
Where does JERP get its information?
From the sites themselves and from official public records, such as company registers. Every field in an answer names its source. The free tier works on sites you name or sites already in the JERP open index; it does not crawl arbitrary URLs on demand.
Do I need an account or a credit card?
No. The free tier gives you a 1-hour API key without an account or a card. Signing in with Google or GitHub is optional: it gets you a persistent key, watchlists, and a note when Pro and Enterprise open.
Is Pro or Enterprise live yet?
Not yet. The free tier is what works today. Pro ($5/mo) and Enterprise ($25/mo) are planned and not billed; sign in free and we will email you when they open.
How do I use JERP in Claude?
In Claude, open Customize → Connectors → Add custom connector, name it JERP, paste the public HTTPS URL ending in /mcp, and choose Sign in. Then, in a chat, open + → Connectors and turn JERP on. Ready-made prompts are in the docs. Localhost URLs do not work — Claude’s servers must reach your host.
How do I use JERP in Cursor?
Copy the mcp.json snippet from the docs into .cursor/mcp.json (project) or ~/.cursor/mcp.json (global), replace jerp_YOUR_PUBLIC_KEY with your key, restart Cursor, and ask it to use the jerp server.
Can I use JERP inside ChatGPT?
On ChatGPT Business, Enterprise, and Edu, yes: add a custom MCP app on the web (developer mode → Apps → Create, URL ending in /mcp, Sign in with your key). ChatGPT Free cannot attach a custom MCP server, and Pro developer mode is read-only, which is not enough. If you run your own agent, call the REST API instead.
How do I call JERP from my own agent?
Call the REST API with Authorization: Bearer and your key, or attach the streamable HTTP MCP endpoint at /mcp. The full contract for agents is at /llms-full.txt and /docs.md.
Why only 24 sites per call?
The cap keeps each round trip fast and cheap enough for a live chat. If you need more, your agent simply makes another call.
Stop guessing. Start checking.
Ask your agent one real question. Get a verdict, not a guess.
Try Free