Jev and MCP
An MCP server turns a Jev call into something your agent can ask for itself. 15 of them are catalogued here, from a plain wrapper around the TypeSafe API to a dispatcher that picks tools with no language model in the loop at all.
Updated 22 Sept 2026 · by Made with Jev
In short
- MCP gives the agent the call. Your code keeps its own calls for everything it already controls.
- The plain wrappers — typesafe-mcp and hermes-typesafe-jev — put Choice, Score and Noul into Claude Code, Claude Desktop, Cursor and Codex.
- The interesting ones go further: tool dispatch with no LLM, browser control, ranked search.
- Every server here is open source and linked to its repository.
The three shapes
Almost every server here is one of three things, and the difference decides which one you want. A wrapper exposes Jev’s own question types, and your agent decides when a judgment is worth asking for. A tool exposes a task — search, screening, ranking, browsing — and keeps Jev inside it, which is usually closer to what an agent actually needs. A router inverts the arrangement: Jev reads the available tools and picks one, which is cheap enough to run on every keystroke and, in one case, removes the language model from dispatch entirely.
| Server | What it exposes |
|---|---|
| typesafe-mcp | Choice, Score and Noul inside Claude Code, Claude Desktop and Codex |
| hermes-typesafe-jev | The same three, as first-class MCP tools |
| jev-studio v0.2.0 | One pip install: MCP tools for all three, plus a CLI |
| mcp_jev | Ready-made question packs, run locally with your own key |
| jev-cli | Jev from a shell, answering with an exit code |
| jev-mcp | Fact checking, prompt-injection screening and semantic ranking |
| jev-search-mcp | Web search where Jev picks the sources and ranks the results |
| Jev Sift | Classify first, read selectively: an agent plugin and MCP tool |
| jevgrep | Code search by behaviour rather than by string |
| AskJev | A real browser, driven click by click, with a guard on irreversible actions |
| jev-mcp-dispatcher | Tool selection and argument extraction with no LLM in the loop |
| jev-tool-router | Picks the tool for Codex instead of listing every one of them |
| maza | A local MCP gateway where Jev handles tool discovery |
| A side panel that drives any site's WebMCP tools | Picks the relevant page tool on every keystroke |
| jev-mcp for coding loops | Jev on the coding loop in Cursor, Codex and any MCP client |
Jev itself, as tools your agent can call
The thin wrappers: Choice, Score and Noul exposed as MCP tools, so Claude Code, Claude Desktop, Cursor or Codex can ask for a typed judgment mid-task. Start here.
GitHubTools and apps
typesafe-mcp
An MCP connector that gives your agent direct access to Jev.
Mostafa
62Andy Holst
Utkarsh Upadhyay ⚒️
@utk2103
just shipped jev-studio v0.2.0 @typesafeai Jev in one pip install - MCP tools for Choice / Noul / Score - `jev` CLI now with dry-run provenance - ready-made prompt libraries + slash commands for every cookbook - Claude Code + Codex plugin manifests pip install jev-studio
XTools and apps
jev-studio v0.2.0
7Stars
Shaharia Lab
One job, done with Jev behind it
Servers that do not expose Jev at all. They expose a task — search, screening, ranking, browsing — and use Jev inside it, which is usually what an agent actually wants.
GitHubTools and apps
jev-mcp
An MCP server with claim checks, screening and ranking built on Jev.
Joey Kudish
70MUKi
GitHubTools and apps
Jev Sift
Classify first, read selectively: an agent plugin and MCP tool.
Kush Bhuwalka
~1628Stars
Nassim Arifette
Ranjan
@manofsteel3129
built askjev on typesafe jev for all-site navigation with claude you talk to claude in plain english and askjev runs your real browser on any site. jev decides every next click — open pages, switch tabs, scroll feeds, fill forms, run multi-step goals without you babysitting the DOM. mcp server + chrome/brave extension. auto-connect once, then stay in chat while the browser moves. claude handles the conversation. jev handles the decision on each step. askjev is the hands on the web. install: load the extension → paste your typesafe key → auto-connect → restart claude → talk example: use askjev, open http://x.com and scroll my feed and find the best posts http://github.com/ranjan2829/AskJev npx -y askjev-mcp
GitHubAgents and browsers
AskJev
Jev deciding which tool the agent gets
The inversion: instead of the agent calling Jev, Jev reads the available tools and picks one. Cheap enough to run on every keystroke, and in one case it removes the language model from dispatch entirely.
GitHubAgents and browsers
jev-mcp-dispatcher
Plain-language MCP tool calls with no LLM at all, only Jev.
abhishekashokvkumar
4Stars
jackbarunz
Sarah Drasner
@sarah_edo
🎇 I made a small Chrome extension: a side panel that drives any site's WebMCP tools with Jev! When you type, on every keystroke it picks the relevant page's tool, fills in the arguments, and tells you how sure it is. Here it is grocery shopping:
XAgents and browsers
PickA side panel that drives any site's WebMCP tools
GitHubAgents and browsers
jev-mcp for coding loops
An MCP server that puts Jev on the coding loop in Cursor and Codex.
burnigtm
MCP, the skill, or a direct call
Three ways to put Jev next to an agent, and they solve different problems. A direct call is right when your own code makes the decision: it is one HTTP request, and it is what every production build in this directory does. The agent skill is right when you want a coding agent to write Jev calls correctly into the code it produces. MCP is right when the agent itself should be able to ask for a judgment mid-task — which tool to use, whether a step is safe, which of these results matters.
If you are still deciding what to point any of them at, the use cases group every build here by job, and how to use Jev has the first call.
Common questions
- What is a Jev MCP server?
- An MCP server that exposes Jev to an agent as a tool. Instead of you writing code that calls the TypeSafe API, the agent itself asks for a typed judgment while it works — which option, is this true, how does this score.
- Should I use MCP or call Jev directly?
- Call it directly from your own code when the decision is part of a program you control; the call is one HTTP request and adding a protocol between you and it buys nothing. Use MCP when the thing making the decision is an agent, and you want it to be able to ask.
- What is the difference between the skill and an MCP server?
- The TypeSafe agent skill teaches a coding agent to write programs that call Jev. An MCP server lets the agent call Jev itself, at the moment it needs a judgment. Many people run both.
- Can an agent run on Jev with no LLM?
- For tool dispatch, yes. jev-mcp-dispatcher reads an MCP server's tool signatures at runtime and uses Choice and Noul to pick the tool and pull its arguments out of your sentence, with no general-purpose model anywhere in the path.
Made with Jev is independent and not affiliated with TypeSafe AI. Every figure on this page is the one its author published, linked to where it can be checked.