mr-remon219/search-boost: A Multi-Engine Search and Evidence Synthesis Tool for AI Coding Agents

Introduction¶

The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When an agent needs to obtain information from external sources, traditional point searches are often inefficient and produce messy data structures. This plugin provides a unified solution aimed at addressing the fragmentation of web search and evidence collection in AI programming agents.

What It Is¶

This is a web tool plugin maintained by developer Mr-remon219, licensed under the MIT license. Through a shared core runtime, it implements features such as multi-engine search, web content extraction, X/Twitter community intelligence gathering, and Jev intent-guided search. It supports a unified MCP server, Pi extensions, and the DeepSeek Harness (DSH) Bundle, and can adapt to different development environments.

Core Features¶

This plugin provides the following core capabilities:

1. Multi-engine parallel fusion (fused_search)
   Supports simultaneous queries to multiple search engine providers. It includes a built-in free engine pool that requires no API Key (such as Bing, DuckDuckGo, and Yahoo) and paid pools that support API Keys (such as Tavily and Brave). It also provides automatic cross-engine URL deduplication, domain routing, and relevance re-ranking.

2. Web content extractor (fetch_page)
   Prefers fetching content from the origin site to reduce latency, with a curl-compatible same-route fallback and Jina Reader as an alternative. It automatically strips CSS, JS, and ad noise, supports focused context paragraph extraction via the focus parameter, and includes memory caching and size limits.

3. X/Twitter community intelligence (x_search)
   Retrieves public posts, user timelines, and discussion threads through the official xAI API or an anonymous fallback channel. It infers UTC timestamps based on verifiable Snowflake IDs and applies author/date filters only when metadata can be verified. Note: coverage may be incomplete or outdated.

4. Jev intent-guided search (adaptive_search)
   This is an experimental feature. By providing the complete question and required research intent, the system automatically selects fixed ranking presets and community branch filtering strategies. It supports filtering search results by fixed safety and prototype values (3/4/5).

5. Native multi-agent parallel research
   Bundles the search-boost and search-boost-parallel-research skills. On hosts that support sub-agents (such as Cursor, Claude Code, Pi, and DSH), tasks can be distributed to parallel searchers (for evidence collection) and summarizers (pure synthesis), supporting fast and complex two-wave workflows.

6. Unified ecosystem core
   Provides a host-agnostic core runtime, drives the standard MCP server, and offers native extensions and Cordis plugin package adaptation for Pi and DSH.

Installation and Enablement¶

Install it through the official Bundle mechanism or by following the source code compilation instructions. This tool depends on a Node.js environment and requires >=22.13.

Typical Usage¶

1. Parallel queries: Call the fused_search tool to issue parallel queries to multiple search engines.
2. Low-latency fetching: Call fetch_page; the system first tries to fetch content from the origin site to reduce latency.
3. Social intelligence: Call x_search to use the official API to retrieve public discussions or user timelines on a specific topic.
4. Intent-driven search: When calling a search tool, provide a complete question description and a specific research intent (such as “compare” or “summarize”) to trigger the filtering logic of adaptive_search.

Applicable Scenarios and Caveats¶

• Applicable scenarios: Developers who need to provide web access capabilities to AI programming agents, perform multi-source information verification, or collect intelligence on a specific platform (such as X).
• Security and privacy:
  • The plugin supports starting the free engine pool with zero configuration and does not require an API Key.
  • Sensitive credentials are stored in plain text in a local private configuration file, with file permissions restricted to POSIX 0600 (not statically encrypted).
  • It uses local networks and proxy DNS, with bounded requests and explicit fallback support.
• Limitations:
  • Search coverage may be incomplete, stale, or empty.
  • No guarantees are made about the accuracy or completeness of the results.

Short Ending¶

This plugin provides a mature web toolchain for the DeepSeek Harness ecosystem. If you are building an agent that requires deep information retrieval, you can integrate it through the plugin directory or the GitHub repository.