The free search bridge for LLMs that can't reach the web

What it does

ModSearch is a plugin for AI agent harnesses—DeepSeek Harness, Claude Code, Codex, and others—that adds web search, X search, and page fetching to models with no native internet access. It returns structured JSON evidence complete with source links and an explicit uncertainty field when synthesis is tentative. The default engine is Firecrawl’s keyless tier, which offers a monthly quota with no account or API key required.

The interesting bit

Instead of relying on a single provider, ModSearch treats search as a utility with redundant paths. It automatically fails over across Firecrawl, Antigravity, Tavily, Exa, and Grok for X queries, and supports per-engine key rotation so an exhausted quota simply rolls to the next credential. That design turns a patchwork of free tiers into one continuous search backend.

Key highlights

• Starts free out of the box on Firecrawl’s keyless tier; no signup or credit card is required.
• Reaches X content via Grok Build, covering ground standard web indexes miss.
• Automatic failover and per-engine key rotation keep queries alive when one channel hits its limit.
• Returns structured JSON with source citations and an uncertainty caveat for low-confidence claims.
• Plugs into DeepSeek Harness, Claude Code, Codex, Pi, and OpenCode as a native skill.

Caveats

• X search requires Grok Build and a SuperGrok or X Premium subscription, so it is not free like the rest of the stack.
• Demo runs show searches and fetches taking anywhere from eighteen to sixty seconds to complete.
• The maintainer does not accept pull requests, so bug fixes and features rely on issues or forks.

Verdict

Worth a look if you run DeepSeek, GLM, or similar models inside agent harnesses and need grounded, cited web answers without managing paid API keys. Skip it if your setup already has native browsing or you need consistently fast turnaround.