psearch
Parallel web search CLI and MCP server where Jev scores fetched pages and picks which links to follow next.
Overview
psearch is a web search tool for terminals and agents, distributed as jev-psearch. Parallel Search's free anonymous MCP endpoint supplies the initial results, a local Chromium fetches pages concurrently, and Jev scores the evidence and selects links for a breadth-first exploration queue. Output is readable text with titles, URLs, scores and excerpts, or JSON with full page text, errors and usage measurements.
Each fetched batch (default six pages) goes to Jev in one request: the state contains the successful pages and their outgoing links, with separate questions per page and per link. Depth and page budget are configurable, and seeds can start the crawl from known pages. The same tool runs as a stdio MCP server exposing one search tool, so agents such as Claude Code can use it. A sibling project, jev-bfs, applies the same idea to the Wikipedia game.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this page as evidence for the query?(inferred) | — |
| Yes / no | Should this outgoing link be followed?(inferred) | — |
How it works
- Get initial results from Parallel Search
- Fetch pages concurrently in local Chromium
- Send a batch of pages and links to Jev in one request with per-page and per-link questions
- Rank sources by score and enqueue selected links, depth by depth, within the page budget
- Return ranked sources as text, JSON or an MCP tool result
Strengths
- One Jev request per batch keeps calls few
- Breadth-first queue with explicit depth and page budgets
- Works as a CLI and an MCP server
Caveats
- Needs a TypeSafe key for Jev calls
- Discovery relies on Parallel's anonymous endpoint, which is rate limited
- No licence stated in the README excerpt
Getting started
uv tool install git+https://github.com/komikat/psearch.git && psearch --install-browser, then export JEVON_KEY and run psearch "your query".At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI and MCP server
- Has code
- yes
Use cases and integrations
How curated lists describe it
Parallel web search for terminals and agents, with local Chromium and Jev-guided exploration.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide in psearch?
It scores each fetched page as evidence and decides which links get added to the exploration queue.
How do I use it from Claude Code?
Run claude mcp add --scope user --transport stdio psearch -- psearch --mcp.
What keys are needed?
A TypeSafe key in JEVON_KEY or TYPESAFE_API_KEY; Parallel Search discovery needs no key.
Related projects
Agent skill where Ego Lite drives the browser and Jev resolves semantic choices: targets, filtering, ranking and evidence checks.
ProjectbetaMarkdown + TypeScriptMCP tool that asks Jev which files, web pages or snippets are relevant to a query, so the agent reads selectively.
MCP serverbetaTypeScriptSelf-hosted memory MCP server on Cloudflare Workers where a Jev layer judges memory type, salience and durability and re-ranks recall.
MCP serverprototypeTypeScriptLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptAgent skill and Python library for Claude Code and Codex that reranks candidate lists with one Jev yes/no relevance question each.
ProjectbetaPythonMCP server that scores every search query, result and fetched page an agent sees for relevance, credibility and steering risk.
MCP serverbetaTypeScript