mcp_jev
Local MCP server that runs versioned packs of typed Jev judgments, with a closed tool catalogue and a key configured once.
Overview
mcp_jev is a local MCP server that serves as a judgment layer for agents and harnesses. It runs versioned 'packs' of typed Choice, Noul and Score judgments, reimplementing common ecosystem patterns such as computer-use steps, review pipelines and model routing, and offers run_questions for custom typed judgments when no pack fits. It deliberately exposes a closed catalogue of tools (list, describe and run packs, run questions, ping) and no free-form ask tool, because Jev is not a chat model or an executor.
Each user runs it on their own machine with their own TypeSafe key, configured once in ~/.mcp_jev during install and reused by any agent that attaches the server, without pasting the key into host configuration files. The harness owns observation, tools, strings and stop rules; mcp_jev owns the typed calls. Documentation includes a skill file, a custom-judgment guide, a blind dogfood protocol, a backlog and installation guides for various agents and IDEs.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Pack-defined choice (e.g. next computer-use action, model route)(inferred) | — |
| Yes / no | Pack-defined yes/no checks (e.g. review gates)(inferred) | — |
| Other | Custom typed questions via run_questions | — |
How it works
- Install mcp_jev and configure the key once
- Attach the MCP server to an agent
- Agent lists and runs a matching pack, or builds custom questions
- mcp_jev calls systemOne and returns typed answers
- Harness acts on the answers
Strengths
- Closed tool catalogue avoids misuse as a chatbot
- Key stored once outside host configs
- Reusable versioned packs
Caveats
- Packs cover common patterns only
- Requires a TypeSafe key
- Agent must decide when to call it
Getting started
Follow the install section to set up ~/.mcp_jev, then point your agent's MCP config at ~/.mcp_jev/bin/mcp_jev.At a glance
- Type
- MCP server
- Status
- beta
- Runtime
- local MCP server
- Has code
- yes
Use cases and integrations
How curated lists describe it
Open MCP server to run TypeSafe Jev (System One) packs locally — Choice / Noul…
wh000wh000/awesome-jev-live
Frequently asked questions
Is there a free-form ask tool?
No, only closed pack and typed-question tools.
Where is my key stored?
In ~/.mcp_jev, configured once and reused by attached agents.
What is a pack?
A versioned set of typed questions for a common pattern.
Related projects
DeepSeek Harness plugin that registers a native jev_decide tool and declares its own policy for when the agent should ask Jev.
PluginbetaJavaScriptHermes skill that calls Jev as a typed tiebreaker, returning Choice, Score or Noul probabilities so the agent can act or escalate.
Agent skillbetaPythonPortable decision layer for agent harnesses: Jev picks from host-supplied capabilities while the host keeps permissions and execution.
ProjectbetaTypeScriptRust MCP server for Codex that hands classification, relevance, candidate selection and evaluation to Jev with accept/verify policies.
MCP serverbetaRustJev decision layer for agents: seven MCP tools, an embeddable library and escalate-only Claude Code hooks at harness boundaries.
MCP serverbetaTypeScriptPython router that has Jev classify Grok Bot requests before costly work, returning actions like reuse_cache, stop_retry or ask_human.
Agent skillprototypePython