liustack/modsearch

ModSearch – Free web‑search plug‑in for LLM chat apps

What it is

• A plug‑in (called a skill) that gives large language models (LLMs) such as DeepSeek, GLM, Claude Code, Codex, Pi, OpenCode, etc. the ability to search the public web, fetch a single page, or query X (Twitter) without the model having its own built‑in browser.
• It works by routing the request through a chain of third‑party search back‑ends (Firecrawl, Antigravity CLI, Tavily, Exa, Grok Build, or a local fetcher). The default is Firecrawl’s key‑less tier, which provides 1 000 free credits per month with no signup.
• If one engine fails or runs out of quota, the next one is tried automatically (fail‑over). Multiple API keys can be supplied per engine and will be rotated on errors.

How you use it

1. Install – one line from the chat interface or the command line, e.g.

   npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modsearch@5.10.6
   # or
   npx skills add liustack/modsearch
   

2. Configure (optional) – add API keys for Tavily, Exa, Firecrawl, etc. in ~/.modsearch/config.json or via environment variables (TAVILY_API_KEY, …). You can also enable/disable individual engines.
3. Chat – simply ask the model a question that needs up‑to‑date information or paste a URL. The plug‑in detects the intent, picks an engine, runs the search/fetch, and returns a JSON‑structured answer with sources, confidence notes, and an optional uncertainty field.

Key features

• Free‑by‑default: works out of the box on Firecrawl’s key‑less tier (1 000 credits/month). Other engines are also free or have low‑cost free quotas.
• Automatic fail‑over & key rotation – seamless switching between engines when a quota is exhausted or a request errors.
• X (Twitter) search via Grok Build, letting the model see content that typical web indexes miss.
• Works across many LLM front‑ends – DeepSeek Harness (dsh), Claude Code, Codex, Pi, OpenCode, etc.
• CLI & config docs – full command‑line interface (modsearch), detailed configuration guide, output schema, and security notes (SSRF guards, private‑network blocking).

Supported back‑ends

Engine	What it does	Free tier	How to enable
Firecrawl (default)	Web search + page fetch	1 000 free credits/month, key‑less	Works immediately
Antigravity CLI	Web search + fetch (better synthesis)	Free, browser sign‑in	agy install & sign‑in
Tavily	Web search	1 000 free credits/month	modsearch config set tavily.apiKey <key>
Exa	Web search	$10/mo recurring credit (~1 400 searches)	modsearch config set exa.apiKey <key>
Grok Build	X (Twitter) search	Uses your SuperGrok/X Premium	Install grok & sign‑in
local	Page fetch only	Built‑in	No extra step

Typical use cases

• Get a concise, sourced summary of a blog post by sending its URL.
• Ask “What’s interesting in AI today?” and receive several recent, sourced news items with a confidence note.
• Retrieve the maintenance status of a software project by fetching its release page.
• Pull the latest tweets on a topic when the model’s own knowledge is stale.

Documentation & support

• Installation guide, CLI manual, configuration reference, output schema, security considerations, and troubleshooting are all in the repo.
• Issues are handled via GitHub; community updates are announced on X (@liustack).
• The project is MIT‑licensed and maintained by a single author; pull requests are not accepted, but forks are allowed.

Bottom line ModSearch is a practical, free‑first plug‑in that restores live web access for LLMs that otherwise lack it, using a smart engine‑fallback system and minimal setup. It is especially useful for developers building AI agents or chat‑based tools that need up‑to‑date information without managing their own search infrastructure.