What it is
Viorant, a start-up that builds deployment infrastructure for AI agents, runs TypeSafe's Jev decision model as a remote MCP server at https://jev.viorant.io/mcp. You sign in with a Viorant account over OAuth, using standard discovery and dynamic client registration; no TypeSafe account or key is needed, and the agent does not have to run on Viorant. A docs pull request merged on 7 October 2026 took the connector out of tester-only access.
It has six tools. Three map to Jev's question types: jev_noul (probability that each statement is true), jev_choice (one option from up to 255, with a probability for each) and jev_score (a position on an ordered rubric of 2 to 10 levels). Three are Viorant's own, built on those primitives: jev_verify checks claims against evidence, jev_screen flags prompt injection and empty or irrelevant content in outside text, and jev_find ranks candidates against a plain-language query without embeddings or an index.
Setup is one command in Claude Code (claude mcp add --transport http jev https://jev.viorant.io/mcp), with documented configurations for Cursor, VS Code, Windsurf, Codex CLI and Gemini CLI. The server code is not published.
Install
https://jev.viorant.io/mcpOur review
Viorant's endpoint is the only hosted way to get Jev with ready-made guard tools rather than raw primitives. jev_screen, jev_verify and jev_find turn Jev's three question types into checks an agent can call between steps, and the connector reaches any client that speaks remote MCP with OAuth, including ones that cannot hold an API key in an environment variable.
The convenience has a price. TypeSafe lists Jev at $0.042 per million input tokens; Viorant charges 100 credits per million, and its smallest top-up sells 10,000 credits for $10, so the effective rate is about $0.10 per million. In absolute terms that is still tiny: the blog's worked jev_verify call used 148 input tokens. But it is a markup on a commodity, billed in a private currency at a 'launch rate' that Viorant can change.
The limits are now published in Viorant's docs: 120 calls a minute, 20,000 a day and 24,000 input tokens per call, all returned as tool-error text rather than HTTP codes. The error table also includes 'Jev is full for free accounts right now', so free-tier capacity is not guaranteed. The blog describes the free 25 million tokens as a launch offer for the first 1,000 users, while the product page presents it as the standard free plan.
On data, Viorant's privacy policy says it does not use prompts or outputs to train any model, and content routed to a third-party provider, here TypeSafe, falls under that provider's terms. The retention periods in the same policy are still printed in square brackets, as in a draft. The tools are closed source, so teams that need to audit what jev_screen actually asks Jev will find the open-source Jev MCP easier to defend.
Where it earns its place
A research engineer running a browsing agent in Claude Code
Stop fetched pages from injecting instructions into the agent
The agent passes each page's raw text to jev_screen with the task purpose and skips anything flagged. A 3,000-token page costs 0.3 credits; at 120 calls a minute the screen will not throttle a single agent.
A docs lead at a 30-person SaaS company
Check release notes against the changelog before they merge
A coding agent sends the draft's claims and the changelog to jev_verify and blocks the merge on any contradicted claim. Each check is a few hundred input tokens, so a month of releases fits easily inside the free 2,500 credits.
A head of support evaluating Jev for ticket triage
Trial Jev from claude.ai before asking engineering for a TypeSafe integration
Add the URL as a custom connector and score a sample of tickets with jev_score. Zero setup cost, but production triage at volume is cheaper through TypeSafe's API directly at $0.042 per million tokens.
Before you commit
- Setup
- Add https://jev.viorant.io/mcp to the client and sign in with a Viorant account in the browser; Viorant documents Claude Code, Cursor, VS Code, Windsurf, Codex CLI, Gemini CLI and its own Hub and vio CLI. First tool call takes a few minutes, most of it creating the account.
- Integrations
- Six tools, all read-only: jev_noul, jev_choice, jev_score, jev_verify, jev_screen and jev_find. Every result carries your remaining credit balance in an allowance_remaining field.
- Data and privacy
- Prompts, context and evidence go to Viorant and on to TypeSafe. Viorant's privacy policy says it does not train on prompts or outputs; its retention periods are shown as bracketed placeholders, and TypeSafe's own terms apply to what it receives.
- Cost at scale
- 100 credits per million input tokens; credit packs run from $10 for 10,000 to $500 for 575,000, so roughly $0.087 to $0.10 per million. An agent screening 1,000 pages of 3,000 tokens a day spends 300 credits a day, about $9 a month.
- Limits
- 120 calls per 60 seconds, 20,000 calls a day and 24,000 input tokens per call. Text only, no explanations, and free accounts can be turned away when capacity is full.
What's good
- Nothing to install; standard remote MCP with OAuth and dynamic client registration
- No TypeSafe account or key needed
- Injection screening, claim checks and ranking alongside the raw question types
- Rate, daily and per-call limits documented
What's not
- About $0.10 per million input tokens against TypeSafe's $0.042
- Adds a second company that sees your prompts and evidence
- Closed source; tool behaviour cannot be audited
- Free accounts can be refused when Jev is at capacity
Best for
People who want to try Jev from Claude, Cursor or another remote-capable client without managing keys or installing anything.
Alternatives
Jev on Viorant: common questions
- Is Jev on Viorant worth it?
- Jev's guardrail tools with nothing to install, at roughly two and a half times TypeSafe's own price and with a second company reading every prompt. We rate it 3.8 out of 5.
- How much does Jev on Viorant cost?
- 100 Vio Credits per million input tokens at the launch rate; output is not billed. Free plan: 2,500 one-time credits. Top-ups from $10 for 10,000 credits, which works out at about $0.10 per million Jev input tokens.
- What are the downsides of Jev on Viorant?
- About $0.10 per million input tokens against TypeSafe's $0.042; adds a second company that sees your prompts and evidence; closed source; tool behaviour cannot be audited; free accounts can be refused when Jev is at capacity.
- Who is Jev on Viorant best for?
- People who want to try Jev from Claude, Cursor or another remote-capable client without managing keys or installing anything.
- How much does Jev on Viorant cost for a team or in production?
- 100 credits per million input tokens; credit packs run from $10 for 10,000 to $500 for 575,000, so roughly $0.087 to $0.10 per million. An agent screening 1,000 pages of 3,000 tokens a day spends 300 credits a day, about $9 a month.
- Is Jev on Viorant safe for company data?
- Prompts, context and evidence go to Viorant and on to TypeSafe. Viorant's privacy policy says it does not train on prompts or outputs; its retention periods are shown as bracketed placeholders, and TypeSafe's own terms apply to what it receives.
- What are the best Jev on Viorant alternatives?
- The closest alternatives we have reviewed are Jev MCP, Jev connector for AnythingMCP and System One Connector.
Answers checked Oct 23, 2026.