Can I install this from npm like the other MCP servers?

No. The install path is a git clone, bun install --frozen-lockfile and scripts/install.sh, followed by exporting TYPESAFE_API_KEY, and it requires Bun 1.3.14 or newer. Built with Effect and TypeSafe's SDK, it runs as a stdio MCP server, so from your side the work is registering it in a harness config rather than adding a dependency.

Where does my API key end up?

Nowhere on disk, deliberately. The jev install commands do not store TYPESAFE_API_KEY; they set JEV_HARNESS and leave it to the harness process to hand the key to the MCP server. The key is read at call time and never logged, which means a misconfigured harness fails at call time with an auth error rather than at install time.

Is any of my source code going to end up in the event log?

The design says no, with limits worth reading. Rank rejects common code, diff and transcript patterns, redacts credentials, and never logs query or candidate text; verification redacts credentials and never writes claim or evidence text. The README also calls those pattern checks conservative rather than a complete classifier, and typesafe_ask state is whatever the caller passes, so redact it on your side.

How do I wire it into my agent, and does it break my existing MCP config?

Run jev install opencode, jev install pi, jev install omp or jev install claude-code. Each writes to a user-level harness config, preserves unrelated MCP settings and sets JEV_HARNESS automatically. The pi path additionally installs pi-mcp-adapter and requires the Pi CLI and network access. docs/mcp.md covers the configuration paths and the tool schemas.

What are the three tools and what does each do?

typesafe_ask makes structured judgments on caller-provided state, typesafe_rank orders supplied candidates by relevance, and typesafe_verify checks claims against evidence you provide. Calls go to the TypeSafe System One API and are recorded in a local event log.

What runtime does the toolkit require?

Bun 1.3.14 or newer to run the application and CLI, installed by a git clone, bun install --frozen-lockfile and scripts/install.sh with TYPESAFE_API_KEY exported. The Oxlint fixture suite separately needs Node 22.18 or newer because that test API is Node-only.

Are the tests online, and do they need a key?

They are offline, using fake transports, temporary files and localhost only. TYPESAFE_API_KEY is read at call time and never logged.

CommunityJev Agents & MCP

Jev Toolkit

A small stdio MCP server that exposes three Jev-backed tools to coding agents: structured judgments, ranking supplied candidates by relevance, and checking claims against evidence you provide.

See highlights

Open github.com

Stars: 1
Language: TypeScript
License: MIT
Published by: Community
Author: jbt95
Last verified: Oct 2026

jbt95/jev-toolkit

User repository on GitHub

View on GitHub

MCP-first toolkit for TypeSafe/Jev — the System One decision model. One stdio server (jev mcp) serves any MCP-capable harness, backed by one local event log and Prometheus impact metrics you can scrape into your own Grafana.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A small stdio MCP server for TypeSafe/Jev with three tools: typesafeask for structured judgments, typesaferank for ordering supplied candidates by relevance, and typesafeverify for checking claims against supplied evidence. Calls use the TypeSafe System One API and are recorded in a local event log. Built with Effect and TypeSafe's SDK. Bun >= 1.3.14 is required. Register Jev in a user-level harness config with:

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Three tools: typesafe_ask for structured judgments, typesafe_rank to order supplied candidates, and typesafe_verify to check claims against evidence.
Register it in a user-level harness config with jev install opencode, pi, omp or claude-code, preserving unrelated MCP settings.
Install is a clone, bun install --frozen-lockfile and scripts/install.sh, then export TYPESAFE_API_KEY; Bun 1.3.14 or newer is required.
Rank rejects common code, diff and transcript patterns and redacts credentials, and verification never writes claim or evidence text to the log.

Watch out

There is no published package: the documented install is a git clone, a frozen-lockfile Bun install and scripts/install.sh. The install commands deliberately do not store TYPESAFE_API_KEY, so the harness process has to provide it to the MCP server, which is an easy thing to get wrong when registering by hand. The pi target installs pi-mcp-adapter and needs the Pi CLI plus network access. The README states its own privacy limits plainly: rank's pattern checks are conservative rather than a complete content classifier, and typesafe_ask state is caller-provided, so you must redact it yourself. Calls go to the TypeSafe System One API and are recorded in a local event log. Tests are offline, using fake transports, temporary files and localhost, and the Oxlint fixture suite needs Node 22.18 or newer even though the application and CLI run on Bun. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

duet-agent

Agent harness for work that outlives one chat, with resumable turns, routed state machines, and observational memory across sessions and processes.

Jev Agents & MCP#community#jevbest#catalog-import

Community

dzhng/duet-agentTypeScript

skilled-agent-harness_spec-driven-loops

Assistant framework that stores specs, memory, and specialist agents as plain markdown files in your repo for Claude Code, Codex, Opencode, and similar runtimes.

Jev Agents & MCP#community#jevbest#catalog-import

Community

MichelKerkmeester/skilled-agent-harness_spec-driven-loopsTypeScript

pi-jev-skill-picker

Pi extension that strips the full Agent Skills catalog from every request and loads only the skills a Jev ranking call judges relevant.

Jev Agents & MCP#community#jevbest#catalog-import

Community

safzanpirani/pi-jev-skill-pickerTypeScript

jevgrep

102

Behavioral code search that asks Jev to score repository fragments for a plain-language question and returns exact excerpts with paths and lines.

Jev Agents & MCP#community#jevbest#catalog-import

Community

nassim-arifette/jevgrepTypeScript

Multi-Agent

Local engine with TypeScript and Python SDKs that runs Claude Code and Codex sessions as a warm team with a durable mailbox.

Jev Agents & MCP#community#jevbest#catalog-import

Community

masonlee39/Multi-AgentTypeScript

oh-my-harness

Single-command harness that turns a plain-language stack description into synced docs plus blocking hooks for Claude Code, Codex, and Pi, with deterministic presets or one Jev chooser call.

Jev Agents & MCP#community#jevbest#catalog-import

Community

kyu1204/oh-my-harnessTypeScript

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai