JevScout
Toolkit where Jev condenses large MCP responses for Claude Code and maps Codex requests to reviewed, fixed repository operations.
Overview
JevScout (yctimlin) uses Jev to cut time and agent overhead in two workflows. For Claude Code, a PostToolUse hook on MCP tools semantically ranks large responses, keeps the relevant source text verbatim and offers exact local recovery of the rest through a jevscout command. The agent receives a condensed packet instead of a notice to read a saved file in chunks, and the user sees a one-line note such as 'kept 5.9 KB of this 207 KB result'. If Jev is unavailable or uncertain, the original response passes through unchanged.
For Codex, a library for applications that drive Codex app-server sessions maps requests to reviewed, fixed repository operations. Host code authorises, executes and verifies the chosen operation and records the result in the conversation; unsupported or uncertain requests continue through normal Codex handling. It requires Node 24+ and a TypeSafe key stored in the system credential store, adds no skills or always-on context, and includes a doctor command. It is independent software, not an official TypeSafe product.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this part of the MCP response to the agent's current task?(inferred) | — |
| Choice | Which reviewed repository operation does this Codex request map to (or none)?(inferred) | — |
How it works
- MCP tool returns a large result in Claude Code
- Hook asks Jev to rank parts; relevant text is kept verbatim
- Agent receives a condensed packet and can recover exact text with jevscout output
- For Codex, a request is matched to a reviewed operation; host authorises, executes and verifies
- Uncertain cases pass through unchanged
Strengths
- Large context savings on MCP results
- Verbatim text and exact recovery, not summaries
- Safe pass-through when Jev is missing or uncertain
Caveats
- Requires Node 24+ and a TypeSafe key
- Codex part is a library for app-server integrators
- Independent project
Getting started
claude plugin marketplace add yctimlin/JevScout && claude plugin install jevscout@jevscout, then /plugin configure jevscout@jevscout to add your key.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Claude Code plugin and Node library
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev decides what reaches your coding agent: verbatim, recoverable context for Claude Code and reviewed operations for Codex. npm i -g jevscout
logicrw/awesome-jev-projects
Frequently asked questions
Does it summarise MCP results?
No, it keeps relevant source text verbatim and lets the agent recover the rest exactly.
What happens without a key?
Results pass through unchanged and the notice says so.
What does the Codex part do?
It maps requests to pre-reviewed repository operations that host code authorises and verifies.
Related projects
Gives an agent one tool that uses Jev to search a nested capability tree and return only the documented commands a task needs.
ProjectprototypeTypeScriptLocal Python MCP service that keeps the skill catalog out of the agent's context and lets Jev pick which SKILL.md files to load.
MCP serverbetaPythonPi extension that strips the inlined skill list and asks Jev which single skill, if any, to load per turn.
Agent skillbetaTypeScriptOpen-source Jev skills for Claude Code, Codex, Cursor and Grok: one client to gate, route, guard, verify, triage, review and act.
ProjectbetaTypeScriptLocal gateway and MCP tools that let Jev predict which tool a coding agent should call next, from suggestions to enforced choices.
MCP serverbetaTypeScriptPre-message hook that keeps or drops SKILL.md files before the first model call, using local overlap or Jev keep/drop judgments.
HookbetaPython