User repository on GitHub
Typed AI decisions for Codex, Claude and any MCP client, powered by TypeSafe Jev. Classify, score and evaluate with one generic tool.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Agents & MCP
Jev MCP Server — mattioofr
An MCP server that gives Codex, Claude and other clients two tools for typed decisions: jev_evaluate mixes noul, choice and score questions in one call, and jev_classify batches up to 100 items against defined classes with review thresholds.
- Stars
- 2
- Language
- JavaScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- mattioofr
- Last verified
- Oct 2026
- Languages
- JavaScript
- Platforms
- MCP
- Capabilities
- Open sourceRequires API key
Highlights
- Two tools: jev_evaluate mixes noul, choice and score questions in one call; jev_classify sends one choice question per item for repeated classification.
- A batch holds at most 100 items and 254 named classes plus a required none class; review is recommended on none, below 0.85 top probability or a 0.2 gap.
- Each response reports the actual model, token usage, total latency and attempt count, using native TypeSafe API question shapes.
- Batching is one logical evaluation per batch, not an automatic fallback; per-item isolation means one jev_evaluate call each and more quota.
- Jev evaluates only the content supplied: it does not fetch URLs, search, explain or act, and confidence is not proof of accuracy.
Watch out
Grounded in the truncated repo README and GitHub metadata read on 2026-10-15; the captured text cuts off during install-from-source, so setup beyond the clone command is not described here. The project is independent and not affiliated with TypeSafe, and the 0.85/0.2 review thresholds are documented as starting values to calibrate.
Questions about this entry
Short answers drawn from the linked project or page.
It gives Codex, Claude and other MCP clients a tool for typed decisions with TypeSafe Jev: classify documents, triage tickets, score relevance, review content against a rubric, or ask several independent questions in one call and get structured answers instead of generated prose. It is independent and needs no database or backend.
Noul for a yes/no judgment returning a probability of yes; choice for selecting one of your named options, returning the choice, the distribution and confidence; and score for rating against ordered descriptive levels, returning a fractional score with distribution and confidence. Types can be mixed in one call.
jev_classify takes a purpose, items with stable IDs and raw content, classes with descriptions, and a required none class. It sends one provider request containing one choice question per item and returns each class probability, the selected class and reviewIds. A batch holds at most 100 items and 254 named classes plus none.
By default when the selected class is none, when the top probability is below 0.85, or when the gap to the next class is below 0.2. The README describes these as starting thresholds that should be calibrated with labeled examples before they trigger any automatic actions.
It is one logical evaluation per batch, not an automatic fallback; if isolation matters the agent can call jev_evaluate once per item, which is one evaluation per item and can consume more quota. HTTP 429 or 5xx retries can add provider requests, and attempts plus upstreamCalls report the actual count.
Jev evaluates only the content you supply: it does not fetch URLs, search the web, write explanations or perform actions, and confidence describes the model's distribution rather than proving factual accuracy. The README says to keep consequential decisions with the calling agent or a human reviewer.
Node.js 22+ and a TypeSafe API key, then clone the repository from GitHub. The server is an independent, community-maintained project with no affiliation to TypeSafe and requires no database or application backend. The README also links a French walkthrough and eight reproducible live scenarios with their raw results.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev in front of the HeyGen MCP
X: Playwright CLI + Jev vs Playwright MCP
X: A local MCP server that cuts computer use from 23s to 0.8s. Cartwright built Cortex, a local MCP server where…
X: A side panel that drives any site's WebMCP tools
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
Python MCP server exposing Jev's three official question types plus compare, verify and batch classify, with an installer subcommand that writes the client config and a measured cost per call.
Jev Practices & Patterns#community#awesomejev#catalog-import
Community
wangkuangkuang/jev-mcp-serverPython
wangkuangkuang/jev-mcp-serverPythonwangkuangkuang's Python MCP bridge to Jev: three official question types plus compare, verify, and batch classify, with a one-command installer that writes client config for Claude Code, Pi, Cursor, OpenCode, or Codex.
Jev Agents & MCP#community#mcp#python
Community
More Jev resources like this
Open-source Codex plugin that consults TypeSafe Jev about tools, models, tasks, skills, context, and strategies, diagnoses failed commands, and checks completion claims against evidence. The current package is version 0.4.0.
Jev Agents & MCP#community#catalog-import#logicrw
A Codex plugin that prepares traceable context handoffs: Jev answers one narrow classification question per old, verified read-only tool pair, and code validates the answers, preserves protected content and builds a JSON or Markdown packet.
Jev Agents & MCP#community#catalog-import#logicrw
A portable Agent Skill that teaches coding agents - Claude Code, OpenAI Codex, Cursor and other compatible hosts - to use Jev as a bounded judgment layer, plus an optional MCP bridge and workflows for code, browser research, Blender and Unreal.
Jev Agents & MCP#community#catalog-import#logicrw
A dependency-free MCP server that exposes TypeSafe's Jev to Claude Code, Codex, Cursor, Hermes, and other agents through the ask_jev and list_jev_models tools, so your code gets typed decisions with probabilities instead of prose to parse.
Jev Agents & MCP#community#catalog-import#logicrw
Community
codeia-academy/jev-mcpPython
codeia-academy/jev-mcpPythonA local stdio MCP server that adds a jev_decide tool to OpenCode agents, supporting noul, choice and score questions so an agent can route or gate work on structured answers and probabilities instead of prose.
Jev Agents & MCP#community#catalog-import#logicrw
Code Mode for MCP where the sub-model is a calibrated decision model instead of an LLM: agents get search(query) and execute(code), and Jev makes per-item choices in about 100 ms, benchmarked on MCPToolBench++, LiveMCPBench and When2Call.
Jev Agents & MCP#community#catalog-import#logicrw
Community
vishiatchoudhary/tooljevPython
vishiatchoudhary/tooljevPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.