Overview & result
This Traditional Chinese-language project brings Jev Search into Claude Code and Codex as an MCP server, CLI, skill and Claude Code WebSearch hook. Jev plans where and how to search and ranks returned links; the agent still reads and reasons over sources.
From the original post
Jev Search for Claude Code and Codex: a function-hook plugin that answers the built-in WebSearch with Jev, plus the same search as an MCP server, a CLI and a skill. Zero runtime dependencies
How Jev fits in the loop
- Interpret a natural-language search request and choose candidate sources, time range and query terms.
- Call Search1API across selected sources and ask Jev to score each returned result.
- Return ranked links and summaries to the agent; fall back to built-in WebSearch on documented hook failures.
How to reproduce
- Install the documented Claude Code plugin or run its npm/MCP package with the required search configuration.
- Submit a plain-language query and inspect selected sources, time window and ranked links.
- Test the documented fallback path when Jev Search is unavailable or returns no usable results.
Why this build matters
An agent-search integration with a clear decision boundary between query planning, retrieval and answer generation.
Limitations
- The plugin depends on Jev Search and Search1API services; it does not generate or verify a final answer.
- Claude Code function hooks are early access; the MCP/CLI mode is the alternative path.