All shelves / Mods / Workflow
jev-search
By mukiwu
jev-search is a Claude Code workflow mod by mukiwu.
Answers Claude Code's WebSearch with Jev Search, and sends each query to your Jev Search instance, jev.s1.dev unless you change it.
Draws
Install
The marketplace name is muki-ai-plugins.
claude plugin marketplace add mukiwu/muki-ai-plugins claude plugin install jev-search@muki-ai-plugins
It runs inside the Claude Code process. The side panel is only what the author's README names. Read the repo before you install, or run claude plugin validate on the plugin folder.
- Version
- 0.3.1
- License
- MIT
- Repo stars
- 7
- Last change
- 2026-10-02
What it can reach
- Draws in the session
- Reaches the network or an outside service
- Sees tool calls
Jev Search answers Claude Code's built-in WebSearch, and the same search is also an MCP server, a command-line tool, and a skill. Needs Claude Code 2.1.271 or later. The README asks you to set CLAUDE_CODE_ENABLE_FUNCTION_HOOKS to 1 in the env block of ~/.claude/settings.json. The official docs say 2.1.287 and later ignore that flag. Restart Claude Code or run /reload-plugins after installing. Intercept starts on. A search that errors, is rate-limited, loses the network, or has nothing left after domain filters falls back to the built-in WebSearch and leaves a dim line in the conversation. After a search, a line names how many results came back, which sources, and how long it took. A short note at the start of the conversation tells the model that WebSearch is now Jev, and WebSearch's description gains a reminder to write the query as one sentence. maxResults defaults to 10. baseUrl defaults to https://jev.s1.dev, and the query goes there. You can change both in /config. The README says a Jev model reads the query, picks sources, a time window, and keywords, then Search1API queries Google, DuckDuckGo, and Yandex, and when needed Hacker News, Reddit, GitHub, X, arXiv, YouTube, Wikipedia, IMDb, and WeChat, and Jev scores each result. What comes back is ranked links and snippets, not a generated answer. The README says it does not search local files, does not fetch page bodies, and does not store query history. A query is cut at 300 characters on a word boundary. The README says the official instance is someone else's bill, with about 10 requests a minute per address, and that heavier use should point baseUrl at your own instance. Upstream development can run at http://localhost:3030 if you point baseUrl there. That is not the default. The command-line tool and the MCP server call that same instance. One session from a checkout is CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude --plugin-dir .
The marketplace name is muki-ai-plugins. The marketplace also holds figma-visual-reviewer, review-tests, lore, session-review, hyday-pet, and world-news. This one installs on its own. Answering WebSearch in the session, and the result line, is why this page files it under workflow. Jev Search, upstream at superagents-lab/jev-search, is a search service in which a Jev model picks the sources and scores the results. jev-cli, claude-jev, jev-rules, and jev-skill-suggestion send their own questions to Jev with a key you supply. This one's settings take no key, and the query goes to your Jev Search instance, https://jev.s1.dev unless you change it. The README says the TypeSafe and Jev names belong to their holders. A failure falls back to the built-in search.
Demo

What it hooks into
prompt.contextA Claude Code eventtool.describeA Claude Code eventWebSearchtool.callAround a tool call, where it can hold oneWebSearch
Read from the on() calls in its hook modules. Open its source under Files to check what each one does.
Files
The author's README and source, 4 files. They load when you open them.
Facts
- Author
- mukiwu
- Category
- Workflow
- Marketplace
- muki-ai-plugins
- Version
- 0.3.1
- License
- MIT
- Repo stars
- 7
- Last change
- 2026-10-02
- Language
- JavaScript
- Files
- 33
