JevTrace
MCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
Overview
JevTrace gives a coding agent the JavaScript or TypeScript code a task needs, within a token budget. It uses the TypeScript compiler to follow calls, types, callers and tests from candidate entry points, and asks Jev to judge the relevance of each piece. It runs as an MCP server for Claude Code, Codex and Cursor, or as a CLI query.
It is designed for questions about one behaviour or area of an existing codebase, such as where something is decided and what else touches it; project-wide requests like 'find bugs' get back suggested per-area subtasks instead of code. It supports .ts, .tsx, .js, .jsx, .mjs, .cjs and Vue components, ES modules and CommonJS, with or without a tsconfig, up to 20,000 files. Jev can be reached through OpenRouter, TypeSafe, Vercel AI Gateway, OpenCode Zen or your own endpoint. The README links a retrieval benchmark on 19 tasks with labels written by the authors, plus an agent benchmark against jevgrep and ttsc.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this code unit to the task?(inferred) | — |
How it works
- Agent calls the MCP tool with a task
- Discovery finds candidate entry points
- Compiler expansion follows calls, types, callers and tests
- Jev judges relevance of candidates
- Relevant code is returned within the token budget
Strengths
- Combines exact compiler structure with semantic relevance
- Token budget bounds the context added
- Multiple provider options
Caveats
- JS/TS only; not Svelte or Astro
- Benchmark labels written by the authors
- Not suited to whole-project review requests
Getting started
npx -y jevtrace setup, then claude mcp add jevtrace --scope user -- npx -y jevtrace.At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- MCP server and CLI (Node 20+)
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
An MCP server that cuts coding-agent token usage by retrieving only the relevant JS/TS code using the TypeScript compiler and Jev.
logicrw/awesome-jev-projectsAn MCP server that cuts coding-agent token usage by retrieving only the…
wh000wh000/awesome-jev-liveAn MCP server that cuts coding-agent token usage by retrieving only the relevant JS/TS code using the TypeScript compiler and Jev.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
How relevant each candidate code unit is to the agent's task.
Which languages are supported?
JavaScript and TypeScript variants plus Vue components.
Which providers can I use?
OpenRouter, TypeSafe, Vercel AI Gateway, OpenCode Zen or a custom endpoint.
Related projects
CLI and MCP server giving coding agents Jev-powered semantic search, focused reads and list picking in about a second for about a cent.
BenchmarkbetaPythonLocal code search for coding agents: ripgrep/BM25 shortlist, then Jev judges each chunk's relevance, served over MCP.
MCP serverbetaRustMCP server that finds relevant code with Jev Choice and Noul checks instead of grep sweeps, and audits diffs against task intent.
MCP serverprototypePython- jevgrep (dzhng)★ 2,671
Code search CLI and agent skill (jg) where Jev judges relevance across folders, files and declarations to return excerpts.
Agent skillbetaTypeScript MCP server giving coding agents fast typed decisions (ship gates, risk checks, file triage, page checks) as accept, review or escalate.
MCP serverbetaTypeScriptRetrieval layer for Codex that finds evidence locally, optionally ranks sanitized excerpts with Jev, and returns bounded, cited evidence.
Projectbeta