jev-toolkit
Small stdio MCP server with ask, rank and verify tools backed by Jev, a local event log and installers for several agent harnesses.
Overview
jev-toolkit is a compact MCP server for TypeSafe's Jev built with Effect and TypeSafe's SDK on Bun. It exposes three tools: typesafe_ask for structured judgments, typesafe_rank for ordering supplied candidates by relevance, and typesafe_verify for checking claims against supplied evidence. Every call uses the System One API and is recorded in a local event log; a listing adds that Prometheus impact metrics can be scraped into your own Grafana.
A `jev install` command registers the server in user-level configs for OpenCode, Pi, OMP and Claude Code while preserving unrelated MCP settings; it does not store the API key, which the harness must supply. Privacy measures include redacting credentials, never logging claim, evidence, query or candidate text, and rejecting common code, diff and transcript patterns in rank inputs, though the README notes these pattern checks are conservative rather than a complete content classifier. Development uses lint, format, typecheck and offline test gates.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | Caller-defined structured judgment via typesafe_ask | — |
| Score | How relevant is each supplied candidate to the query?(inferred) | — |
| Yes / no | Is this claim supported by the supplied evidence?(inferred) | — |
How it works
- Install with Bun and register in a harness via jev install
- Agent calls ask, rank or verify over MCP
- Server calls Jev and returns typed answers
- Event log and metrics record usage without sensitive text
Strengths
- Single server for several coding-agent harnesses
- Careful logging that excludes content and credentials
- Offline test suite and quality gates
Caveats
- Requires Bun >= 1.3.14
- Content pattern checks are not a complete classifier
- Callers must redact ask state themselves
Getting started
git clone, bun install --frozen-lockfile, scripts/install.sh, export TYPESAFE_API_KEY, then `jev install claude-code` (or opencode, pi, omp).At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- MCP server (Bun)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
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.
logicrw/awesome-jev-projects
Frequently asked questions
What tools does it expose?
typesafe_ask, typesafe_rank and typesafe_verify.
Which agents can it be installed into?
OpenCode, Pi, OMP and Claude Code via the jev install command.
Does it log my prompts?
No; it redacts credentials and never logs claim, evidence, query or candidate text.
Related projects
MCP server giving coding agents typed judgment tools (verify, screen, rerank, gate, score) backed by Jev, with auto/review/escalate policy.
MCP serverbetaPythonLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptMCP server and agent skill for designing Jev judgments: question linting, batch evaluation, threshold calibration and reranking.
MCP serverbetaTypeScriptEvidence gate for RAG: a System One model judges whether retrieved excerpts support an answer before the LLM is allowed to speak.
MCP serverbetaPythonCLI and MCP server giving coding agents Jev-powered semantic search, focused reads and list picking in about a second for about a cent.
BenchmarkbetaPythonpip-installable Jev kit: a CLI for verify, screen, classify, extract, route and rerank, plus an MCP server for the primitives.
MCP serverbetaPython