Jev.aitools.fyi

Command Palette

Search for a command to run...

Codex and OpenCode + Jev

jev-judge on the CLI, evaluate setup, and routers that treat Jev as traffic control.

What Codex and OpenCode are

toml). jsonc configuration. md directories.

jev-judge-mcp install

gecm0/jev-judge-mcp documents first-class Codex and OpenCode lines. Register stdio npx -y jev-judge-mcp, install the jev skill, and add typesafe-ai as a companion skill.

codex mcp add jev -- npx -y jev-judge-mcp
opencode mcp add jev -- npx -y jev-judge-mcp

npx skills add gecm0/jev-judge-mcp --skill jev
npx skills add typesafe-ai/skills --skill typesafe-ai

Example Codex MCP table

If you configure by hand, Codex uses a [mcp_servers.jev] table. Prefer environment export for secrets instead of literals when your client version allows it.

[mcp_servers.jev]
command = "npx"
args = ["-y", "jev-judge-mcp"]
# Codex may require env literals in some versions; exporting TYPESAFE_API_KEY in the parent shell is safer.

typesafe-mcp evaluate

The Go evaluate CLI can register the same MCP server with Codex in one step: TYPESAFE_API_KEY=... evaluate setup mcp. OpenRouter keys work when documented in the README.

jev-codex-router

0xNatoshi/jev-codex-router is a Codex Router extension that classifies each turn with Jev and selects model plus reasoning effort, fail-open on errors. It is separate from jev-judge-mcp but shows the same pattern: Jev on the routing edge, native Codex execution behind it.

Using judge well

The bundled jev skill teaches the one-second test: ask factors Jev can answer at a glance, fan out parallel questions, and compose the final decision in code. judge rejects obvious generative requests; supply evidence, not homework for Jev to write.

Environment keys

ai/ (or route through Vercel AI Gateway / OpenRouter where the integration documents it). Export TYPESAFE_API_KEY in the environment for Codex and OpenCode launch environment.

Some MCP clients accept AI_GATEWAY_API_KEY when using gateway-hosted Jev models; follow that integration's README. Never commit keys into repos or paste them into chat logs.

export TYPESAFE_API_KEY="tsk_..."
# Optional pin:
export TYPESAFE_MODEL="jev-latest"

Pitfalls

Restart the client after MCP changes so judge appears in the tool list. TYPESAFE_MODEL pins a version; jev-latest follows TypeSafe's moving alias. Without a key, jev-judge-mcp lists the tool but fails fast with a clear error. com/gecm0/jev-judge-mcp.

Related directory listings

  • Agent toolingPython27
    Per-turn model & reasoning routing for Codex, driven by Jev (TypeSafe System One): picks the model, thinking depth and speed mode for eve...

    by @0xLogicrw

    agent toolingroutingagents
  • Agent toolingGo62
    MCP
    Go CLI and single-binary MCP server exposing TypeSafe judgments to Claude Desktop, Claude Code, and Codex.

    by @0xLogicrw

    agent toolingmcpagents
  • Agent toolingTypeScript4
    MCP
    Brainwires jevwire packages a Jev decision layer for agents: Streamable MCP server, embeddable DecisionModel library, and a Claude Code plugin marketpla
    agent toolingmcpagents
  • Guides and lists
    Preview of fast browser use with Jev and OpenCode's browser CLI.

    by @thdxr

    guidesbrowser-use

Related agent guides

FAQ

Claude Code plugin vs Codex mcp add?
Claude uses marketplace plugins; Codex uses codex mcp add or TOML. The same jev-judge-mcp package backs both via stdio.
OpenCode v2 config shape?
jev-judge-mcp README references opencode.jsonc with mcp.servers.jev and type local. Use opencode mcp add when available so the CLI writes the correct shape for your version.
Can Jev write code for me in Codex?
No. judge is for typed judgments over supplied evidence. Let Codex's primary model generate code.
What is jev/auto in jev-codex-router?
The router exposes a curated model entry that forwards through a local Jev classification server before selecting native Codex models. See the router README for ports and kill switch behavior.

Primary sources

By Rishit