Skip to content

typesafe-mcp (MarkChu-git)

Bun-based MCP server wrapping Jev with compact answers and a code-computed act, review or abstain verdict.

MCP serverbeta

Overview

typesafe-mcp is an MCP server that wraps Jev so any MCP host (Cursor, Claude Desktop, Claude Code, Windsurf, Cline or a custom client) can ask typed questions and get compact, auditable answers. Each answer includes the result, how certain Jev is, and a threshold-gated act, review or abstain decision. That decision is computed in code by the server, not by Jev.

It is designed to use as few tokens of the agent's context as possible. It runs with bunx as a single-file stdio build, and without a key it still lists tools while calls return a configuration error. Settings cover the default model and per-request timeout. It is published on npm with CI.

What the decision model decides

TypeQuestionAnswers
Yes / noAgent-posed yes/no questionyes · no
ChoiceAgent-posed choice—
OtherShould the agent act, review or abstain given this certainty?act · review · abstain

How it works

  1. The agent calls jev_ask with state and typed questions.
  2. The server sends them to Jev.
  3. It computes act, review or abstain from thresholds.
  4. The agent receives a compact answer and verdict.

Strengths

  • Gating verdict computed deterministically in code.
  • Minimal context usage.
  • Works with many MCP hosts via bunx.

Caveats

  • Requires Bun and a TypeSafe key.
  • Community project.

Getting started

bunx typesafe-mcp (add to your MCP host config with TYPESAFE_API_KEY)

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
MCP server (Bun)
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

  • MCP server for TypeSafe AI's Jev: typed yes/no, choice and score answers with act / review / abstain gating.

    hellogumbo/awesome-jev

Frequently asked questions

Who decides act, review or abstain?

The server's code, using thresholds, not Jev itself.

Which hosts are supported?

Any MCP host, including Cursor, Claude Desktop, Claude Code, Windsurf and Cline.

What happens without a key?

Tools list, but calls return a configuration error.

Related projects

  • Claude Code PreToolUse layer where Jev answers YAML-defined questions about each tool call, with 19 hard rules and calibration runs.

    ProjectbetaTypeScript
  • OpenJev for Claude Code: plugin and MCP server giving Claude jev_choice, jev_noul, jev_score and jev_ask tools on the live API.

    MCP serverbetaTypeScript
  • Starter kit that gates every MCP tool call with Kinde identity checks plus Jev risk signals, then allows, holds for approval or stops it.

    MCP serverbetaTypeScript
  • Zero-dependency Python MCP server where Jev gates shell commands, audits patches and resolves decisions, with SQLite caching.

    MCP serverbetaPython
  • Unix-pipeable CLI for Jev and other System One models: ask questions, gate shell commands via exit codes, calibrate thresholds.

    CLIbetaGo
  • Governance proxy between Jev or local Laya and agents that turns probabilities into allow, ask, deny, keep, truncate or drop actions.

    MCP serverbeta