typesafe-as-a-judge
Unofficial Codex and Claude Code MCP plugin giving agents bounded Jev tools for routing, ranking, extraction, verification and judging.
Overview
TypeSafe-as-a-Judge is a dual plugin for Codex and Claude Code that adds a bounded semantic-judgment layer through an MCP server. The agent stays responsible for orchestration, permissions, code changes and side effects; Jev supplies small typed signals the agent uses to route work, rank candidates, verify evidence and decide when to ask for review. It is deliberately not an autonomous authority layer.
Tools include typesafe_route (one item to one known route with a proceed or review recommendation), typesafe_rank (order an already-retrieved candidate list), typesafe_extract (pick a candidate value or no_match, never inventing one), typesafe_verify (support probability for a claim against evidence), typesafe_judge (raw Choice, Score and Noul signals), plus usage summary and session mode controls. It is Apache-2.0 and maintained by one person.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which known route should this item go to? | — |
| Yes / no | Does the evidence support this claim? | yes · no |
| Extract | Which of the found candidate values is the right one, or no_match? | — |
How it works
- The agent calls an MCP tool with an item, candidates or a claim.
- Jev returns typed signals and confidence.
- The tool recommends proceed or review.
- The agent applies its own policy and acts.
Strengths
- Extraction never invents values; it picks a candidate or no_match.
- Proceed or review recommendations support human escalation.
- Usage summary reports calls, tokens and elapsed time.
Caveats
- Unofficial; single maintainer.
- Requires a TypeSafe key and sends data to the hosted model.
At a glance
- Type
- MCP server
- Status
- beta
- Runtime
- MCP server, Codex and Claude Code plugin
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Unofficial community MCP plugin for Codex and Claude Code using TypeSafe Jev for bounded routing, ranking, extraction, verification, and escalation.
hellogumbo/awesome-jev
Frequently asked questions
Does Jev take actions?
No, it only returns typed signals; the agent keeps control of actions.
Can extraction hallucinate a value?
No, it selects from candidates the code found or returns no_match.
Which agents are supported?
Codex and Claude Code.
Related projects
Local memory for coding agents that stores verbatim quotes and has Jev grade memory lines as keep, stale, wrong or unsupported.
ProjectbetaPythonLocal MCP bridge letting Codex hand drafting to Gemini via a Google AI Pro quota, with Jev checking drafts against explicit criteria.
MCP serverbetaVideo summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.
ProjectbetaPythonProof-of-concept MCP tool dispatcher that routes a natural-language command and extracts arguments using only Jev Choice and Noul.
MCP serverprototypePythonIDE and browser extensions to find and install agent Skills and Subagents, with an opt-in Jev check when local detection can't decide.
Browser extensionproductionTypeScriptPortuguese agent skill teaching Claude Code, Cursor and Codex to use Jev correctly, bundled with the full raw research behind each claim.
Agent skillproduction