Skip the install. Get this working in under 2 minutes.
Start a free trial on cloud.anythingmcp.com, add the Jev by TypeSafe in one click, then point your AI client (Claude, ChatGPT, Copilot or Cursor) at the generated MCP endpoint. No Docker, no git clone, zero engineering experience required.
Summary
Give Claude, ChatGPT, Cursor or Codex the Jev decision model: yes/no, classification and scoring with probabilities in about 300 ms. No local install.
Try asking
Example prompts for Jev by TypeSafe
Click any prompt to copy it. Paste into Claude, ChatGPT, Cursor, Gemini, Copilot or OpenClaw to run it against this connector.
Claude is AI and can make mistakes. Please double-check responses.
Jev as an MCP server
Short answer: install the Jev connector on AnythingMCP with your TypeSafe API key, then add AnythingMCP to your AI client as a remote MCP server. Claude, ChatGPT, Cursor, Codex or Gemini can then hand every yes/no, classification or scoring decision to Jev and get back a number with probabilities in about 300 ms. Nothing runs on your laptop, and it works in claude.ai and ChatGPT on the web.
Jev is the first System One model from TypeSafe, released in September 2026. It does not write text. You give it evidence (the state) and typed questions, and it answers each one with a probability, an option or a score. That makes it the right tool for the small judgments an agent makes all the time, and the wrong tool for anything that needs prose.
What Jev answers
| Question type | You define | Jev returns | Typical use |
|---|---|---|---|
Yes/no (noul) | the condition to test | probability that the answer is yes | is this urgent, is this spam, does this reply answer the question |
| Choice | 1 to 255 options, each with a short description | the most likely option, one probability per option, a confidence | routing, labelling, intent, picking the best record |
| Score | 2 to 10 ordered levels, lowest first | a 0-indexed score, one probability per level, a confidence | severity, lead fit, sentiment, quality |
In our tests Jev answered in 220 to 430 ms per call. It is billed only for input tokens (about $0.042 per million); a typical call is 300 to 520 tokens, so a single decision costs about two thousandths of a cent.
Jev vs asking Claude or GPT directly
| Asking the chat model | Asking Jev through the connector | |
|---|---|---|
| Answer | a sentence the agent has to interpret | a number or an option from your list |
| Uncertainty | hidden in wording ("seems fairly urgent") | explicit: 0.52 means unsure, 0.97 means clear |
| Speed | seconds, and it uses the chat model's context | 0.2 to 0.4 s, in parallel for many questions |
| Cost | chat-model tokens | $0.042 per million input tokens, output free |
| Consistency | can change with phrasing | same question, same scale, every record |
The chat model stays in charge: it decides what to ask, reads the probabilities and acts. Jev makes the narrow call.
Set it up
Step 1: Get a TypeSafe API key
Sign in at console.typesafe.ai and create an API key. You pay TypeSafe directly for what Jev processes.
Step 2: Install Jev on AnythingMCP
- Create a workspace on AnythingMCP Cloud (7 days free) or use your self-hosted AnythingMCP.
- Open Connectors, find Jev by TypeSafe, click Install and paste the key into
TYPESAFE_API_KEY. AnythingMCP checks the key with a free call and stores it encrypted. - For clients that use an API key (everything except claude.ai and ChatGPT), create one under MCP Servers → your server → API keys.
Step 3: Connect your AI client
Every client connects to the same MCP server URL, https://cloud.anythingmcp.com/mcp. claude.ai and ChatGPT sign in with OAuth; the others send your MCP API key in the X-API-Key header.
The tools
| Tool | What it does |
|---|---|
jev_playbook | How to use Jev well: question types, state, reading probabilities, confidence thresholds. Free, no API call. |
jev_yes_no | One yes/no question, returns yes_probability |
jev_classify | One choice over options you define, returns choice, probabilities and confidence |
jev_rate | One score over ordered levels, returns score, legend, probabilities and confidence |
jev_ask | Several questions of any type about the same evidence in one call |
jev_list_models | Model names your key can use (jev-latest, jev-preview, pinned versions) |
All tools are read-only: Jev reads the evidence you pass and changes nothing.
What happens when you ask
We gave Claude four support tickets and asked it to triage them with Jev: urgency, owning team and frustration on four levels. Claude read jev_playbook, made one jev_ask call per ticket with three questions each, and set its thresholds before looking at the answers.
| Ticket | Urgent (P yes) | Team | Frustration (0 to 3) |
|---|---|---|---|
| "My payouts have been failing for 3 days and my suppliers are waiting." | 0.88 | billing | 3 |
| "Do you offer a discount for annual plans? Just curious." | 0.02 | sales | 0 |
| "Die App stürzt bei jedem PDF-Upload ab, seit gestern Abend. Wir können nicht arbeiten!" | 0.90 | technical | 3 |
| "Thanks for the great webinar last week :)" | 0.01 | none | 0 |
Each call took 0.3 to 0.4 seconds. Claude flagged the first ticket as the one closest to its "unsure" band, and handled the German ticket without translating it.
Prompts to try
- "Use Jev to decide which of these 20 emails need a reply today, and show me only the ones above 0.8."
- "Classify each of these support tickets into billing, technical, sales or none with Jev, and list the ones where confidence is below 0.6."
- "Rate these five leads from 0 (no fit) to 3 (ideal customer) with Jev against our ideal-customer profile below."
- "Ask Jev whether each of these two customer records describes the same company."
- "Check with Jev whether this reply actually answers the customer's question before I send it."
Jev on your business data
Because Jev runs as one connector among others on the same AnythingMCP server, the agent can fetch records with one connector and judge them with Jev in the same chat: tickets from your helpdesk, leads from your CRM, orders or customer records from your ERP. The data goes to TypeSafe only as the state of the question you asked.
Limits
- Text only. No images, audio or files; turn them into text first.
- Size. Up to 32k tokens for the evidence plus the longest question, 64k per request.
- Languages. English is where Jev is most accurate. German and Italian worked well in our tests; watch the confidence for other languages.
- No reasoning. Jev returns probabilities, not explanations. To audit a borderline answer, ask narrower yes/no questions about the specific evidence.
- Arithmetic and dates. Jev reads literally. Let the chat model do counting and date comparisons and pass the result as a fact.
- Rate limits are set by TypeSafe per account, currently 1,200 requests per minute.
FAQ
Does Jev work in claude.ai and ChatGPT, not just in coding agents? Yes. AnythingMCP is a remote MCP server with OAuth, so you add it as a custom connector in claude.ai or as a developer-mode app in ChatGPT. Nothing has to run on your computer.
How much does it cost? Jev is billed by TypeSafe per input token, about $0.042 per million, output free. A typical decision uses 300 to 520 tokens. The connector itself costs nothing extra: AnythingMCP Cloud has a 7-day free trial, then from €19 a month, and self-hosted AnythingMCP is free.
Does my team need the TypeSafe key? No. The key is stored once, encrypted, on the AnythingMCP server. Team members connect to the server with their own login and never see it, and every call is logged.
Can I self-host it? Yes. AnythingMCP is open source (AGPL-3.0) and runs with Docker Compose. The Jev connector is part of the built-in catalog.
Does Jev understand German or Italian? It handles them, and our German and Italian test cases came back correct and confident. English remains the most accurate; for other languages, keep an eye on the confidence values.
How is this different from the Jev MCP servers on GitHub? Those run locally next to a coding agent and each person configures their own key. The AnythingMCP connector runs on a server, works in web clients such as claude.ai and ChatGPT, is shared by a team with one key and can be combined with your other connectors.
Related
- Connect Jev to Claude
- Connect Jev to ChatGPT
- Triage support tickets with Jev in Claude
- Connect Jev to Cursor
- Connect Jev to Gemini CLI
- Jev by TypeSafe in the marketplace
Was this guide helpful?
