SearchBoost MCP
Multi-engine web search and clean page reading for an agent: keyless parallel search, URL reading, public X posts, MCP and a TUI
- 61 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The tools call third-party search engines and public pages, keys sit in local files as plain text and the installer edits host configurations.
Install
- Works in
Manual install
npm install -g search-boostThen `search-boost` launches the terminal UI for install and host setup.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The tools call third-party search engines and public pages, keys sit in local files as plain text and the installer edits host configurations.
This is third-party code. Review the repository files before installing.
What it does
`SearchBoost` gives agents a web research toolset through one shared core. `fused_search` queries several engines in parallel (a keyless free pool of Bing, DuckDuckGo, Yahoo, Exa-free and AnySearch, and a keyed pool of Tavily, Brave and Exa), removes duplicate URLs and re-ranks for diversity. `fetch_page` reads a public page and returns clean text with no CSS, scripts or ads, extracting PDFs locally. `x_search` finds public X posts, author timelines and discussions. There is an experimental `adaptive_search` that screens for a stated intent, plus `search_stats` and `search_layer`. The core runs as a stdio MCP server for Cursor, Claude Code, Codex, Grok Build and Antigravity, and as native extensions for Pi and a DeepSeek Harness bundle. Install with `npm install -g search-boost` and launch the terminal UI with `search-boost`.
Who it is for. Claude Code, Cursor and Codex users who need multi-engine web search with page reading and no mandatory API keys.
Good fit when
- An agent searches docs and fresh material across several engines at once
- A page and a PDF should be read as clean text by keyword
- A topic discussion on X is gathered through public posts
- Parallel research is delegated to several subagents, searching and summarizing
Not a fit when
- You need pages behind a login, since `fetch_page` is not a signed-in browser
- You need an exhaustive comment sample or total sentiment analysis
- Node.js is older than 22.13 and cannot be updated
Example request
Use search-boost to find the guide to the built-in WebSocket API in Node.js 22 on nodejs.org and MDN, and read the page by the word WebSocket.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
It needs Node.js 22.13 or newer. The free pool relies on public search pages and may be unavailable or limited, and configuration does not guarantee network reachability. Keys are stored in local configs as plain text with `0600` permissions, which is not encryption. `fetch_page` does not read private pages. `x_search` does not promise full comment threads and uses the official xAI API or an anonymous fallback channel. `adaptive_search` is experimental and confirms only the number of selected sources, not verified answers.
How to disable. Disable tools in the Tool switches section of the terminal UI, or remove the MCP entry from the host configuration and run `npm uninstall -g search-boost`. Configs live in `~/.search-boost`.
FAQ
How do I install it?
Run `npm install -g search-boost` and start `search-boost`; the terminal UI offers to connect hosts (Cursor, Claude Code, Codex and others).
Are API keys needed?
No, the free pool works without keys. Keyed engines (Tavily, Brave, Exa) and the official xAI API are optional.
Which tools does the agent get?
`fused_search`, `fetch_page`, `x_search`, the experimental `adaptive_search`, plus `search_stats` and `search_layer`.
How do I turn off a tool?
In the terminal UI open Tool switches, toggle with Space and confirm. Settings live in `~/.search-boost/config/tools.json` and no host restart is needed.
Where does it run?
Cursor and Cursor CLI, Claude Code, Codex, Grok Build and Antigravity through MCP, plus native Pi and DeepSeek Harness.
Discussion
No questions or reviews yet.
Related
MCP serversA local MCP server with no API keys: multi-engine web search, page fetching and document reading tuned for LLMs.
Local web search for agents: searches, reads pages and returns compact passages with sources, saving context tokens
A skill for web search, X search and page fetching for models without web access: no signup or key, the answer is JSON with source links
MCP server with DuckDuckGo search and page text extraction: several URLs can be passed in one call
Keyless web search for agents: a Rust service over HTTP or MCP merges DuckDuckGo, Brave and Yahoo results into one JSON.
Local web search for an agent with no API keys: results are ranked on your machine and only the relevant passages come back
README in short
The README describes the key features, the one core and three adapters architecture, a host table, a quick start, tool switches, a tool reference with boundaries and install from source.