# Jev Filter

> Jev Filter is an open-source MCP server for ai models and llm tooling. Filter search results, code and logs into compact evidence packets for agents. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, Linux, macOS, API
- **Website:** https://apixly-ai.github.io/jev-filter/
- **Repository:** https://github.com/apixly-ai/jev-filter
- **MCP Registry name:** io.github.apixly-ai/jev-filter
- **Categories:** AI models and LLM tooling, developer tools
- **Listing updated:** 2026-11-15

## About

Jev Filter sits between a program's data collection and the main agent. A program gathers search results, code symbols or log events, Jev judges each record against your task, and the agent receives only the selected items, unresolved IDs and review cases. Originals stay recoverable by source ID.

The project publishes a paid test from 2026-10-04 on synthetic inputs: 24 real agent runs showed 91.6 to 97.2 percent less returned tool context, and the filtered agent completed 12 of 12 runs against 11 of 12 with raw context. The authors also report that most measured cells were slower, by 0.1 to 18.4 percent, and say small samples guarantee neither speed nor savings.

It runs through an npm package (@apixly/jev-filter) with a CLI, Python and JavaScript/TypeScript interfaces, and a read-only MCP server. Node.js 22 or later is needed on macOS or Linux, and inference needs a TypeSafe Jev API key whose calls are billable. The project is MIT licensed and maintained by Apixly.

## Features

- **Typed evidence packets:** Returns selected IDs, review IDs and a completeness flag instead of raw output.
- **Recoverable originals:** Every retained record can be fetched back by source ID.
- **Code and log triage:** Judge code symbols by behaviour and separate current failures from recovered log events.
- **Automatic batching:** Batches repeated judgements, with at most 30 concurrent requests and no inference-result cache.
- **Several interfaces:** CLI, Python, JavaScript/TypeScript and a read-only MCP server.
- **Published benchmarks:** Method, per-run data and trade-offs are documented.

## Good for

- Cut noisy search results before they enter an agent's context
- Find which code implements a described behaviour
- Pick the current unresolved failure out of correlated logs
- Screen RAG passages against explicit evidence criteria

## Install

### Claude Code (Terminal)

```sh
claude mcp add jev-filter --env TYPESAFE_API_KEY='<TYPESAFE_API_KEY>' -- npx -y @apixly/jev-filter
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "jev-filter": {
      "command": "npx",
      "args": [
        "-y",
        "@apixly/jev-filter"
      ],
      "env": {
        "TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>"
      }
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "jev-filter": {
      "command": "npx",
      "args": [
        "-y",
        "@apixly/jev-filter"
      ],
      "env": {
        "TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>"
      }
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "jev-filter": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@apixly/jev-filter"
      ],
      "env": {
        "TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>"
      }
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "jev-filter": {
      "command": "npx",
      "args": [
        "-y",
        "@apixly/jev-filter"
      ],
      "env": {
        "TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>"
      }
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "jev-filter": {
      "command": "npx",
      "args": [
        "-y",
        "@apixly/jev-filter"
      ],
      "env": {
        "TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>"
      }
    }
  }
}
```

## FAQ

### What do I need to run it?

Node.js 22 or later on macOS or Linux, the npm package, and a TypeSafe Jev API key for inference.

### Is it free?

The code is MIT licensed, but the README says inference calls are billable.

### Does it make agents faster?

Not reliably. The docs say five of six measured cells were slower, and context reduction is the strongest result.

### Does it work on Windows?

Use WSL or install Python natively, according to the README.

### Is Jev Filter free?

Yes. Jev Filter is open source, so you can use and self-host it for free.

### How do I add Jev Filter to Claude Code?

Run this command in your terminal: claude mcp add jev-filter --env TYPESAFE_API_KEY='<TYPESAFE_API_KEY>' -- npx -y @apixly/jev-filter

### What are the best alternatives to Jev Filter?

The closest alternatives to Jev Filter are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/jev-filter
