JevScout (Claude Code + Codex)

An open-source toolkit that uses TypeSafe's Jev in two coding-agent workflows: condensing large MCP responses in Claude Code, and mapping Codex session requests to reviewed, fixed repository operations with verified execution.

A PostToolUse hook fires for mcp__.* tools, semantically ranks the response, keeps relevant source text verbatim and provides exact local recovery through the jevscout command. If Jev is unavailable or uncertain, the original response passes through unchanged.

Results pass through unchanged. JevScout never writes the key to a settings file, and when a large result passes through because the key is missing or Jev is unavailable, the one-line notice says so.

In the demo, a 169 KB MCP result that Claude Code would read in chunks was answered from a 3-segment verbatim packet in 2 turns instead of 5: 51% less time and 42% less cost on that task, and 38%/26% across all 7 validation tasks.

Add the marketplace and plugin with claude plugin marketplace add yctimlin/JevScout and claude plugin install jevscout@jevscout, configure the key via /plugin configure jevscout@jevscout, then check with jevscout doctor or npx jevscout@latest doctor.

A library for applications that drive Codex app-server sessions: it maps requests to reviewed, fixed repository operations, the host code authorizes, executes and verifies them, and the result is recorded in the conversation. Unsupported or uncertain requests continue through normal Codex handling.

Node.js 24+ and a TypeSafe API key. It adds one hook and one command, no skills or always-on context; plugins cannot add permission rules, so Claude Code asks the first time the agent runs jevscout output. JevScout is independent software, not an official TypeSafe product.