jevex
Single MCP tool that indexes a repo and has Jev rank a shortlist so a coding agent reads the right files and lines instead of grepping.
Overview
jevex (jimmyhealer) is one MCP tool, codebase_investigate, that returns the files and line ranges a coding agent should read. It indexes whatever an agent can already read (Python, TypeScript, Go, Rust, Markdown and more), and when asked a question such as where is session expiry enforced, Jev ranks a shortlist and returns a source-of-truth packet with path, symbol and range plus a sufficiency status. The agent then reads those lines instead of grepping the tree; it still plans and patches itself. The author stresses it is not a second agent and not a thin evaluate wrapper.
On 16 SWE-bench Verified tasks with the same cheap agent (agy with Gemini 3.8 Flash) and a 1,200-second cap, mean time fell from 160 to 69 seconds and the agent bill from $8.74 to $3.13, with the file to patch found in 16 of 16 in both arms; under a 90-second cap only 1 of 16 finished without jevex versus 11 of 16 with it. On a small Claude Code fixture, files read dropped from 6.8 to 2.2. Jev costs about $0.0015 per question. It needs a TypeSafe key and is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How likely is this code location to be the source of truth for the question?(inferred) | — |
How it works
- Index the repository
- Agent calls codebase_investigate with a question
- Jev ranks a shortlist of locations
- Agent reads the returned files and line ranges
Strengths
- README reports 160s to 69s and $8.74 to $3.13 on 16 SWE-bench tasks
- About $0.0015 per question
- Works with any MCP agent
Caveats
- Small sample (n=16)
- Agent still does planning and patching
Getting started
pip install -e ".[mcp]", then `claude mcp add jevex -e JEV_API_KEY=... -e JEV_EXPLORER_REPO=/path/to/repo -- jevex-mcp`.At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- MCP server / CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
One MCP tool that returns the files a coding agent should read.
logicrw/awesome-jev-projects
Frequently asked questions
What does jevex return?
The files and line ranges an agent should read to answer a code question.
How much did it help?
On 16 SWE-bench tasks, mean time fell from 160 to 69 seconds and cost from $8.74 to $3.13.
Does it write patches?
No, it only answers what to read.
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.
BenchmarkbetaPythonDeterministic code-graph context for coding agents on Postgres, AGE and pgvector, with optional Jev or local re-selection of symbols.
MCP serverbetaPythonCLI and MCP server where Jev scores code fragments for a behaviour question and returns exact excerpts with line numbers.
MCP serverbetaTypeScriptMCP server and semantic index that tags shadcn-format components once and matches a plain-language brief to them deterministically.
MCP serverbetaTypeScriptLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptClaude Code plugin that adds Jev as an independent judge for review findings, debugging hypotheses, design options and search hits.
Pluginbeta