modsearch

Models like DeepSeek and GLM have no web access, or a weak one. ModSearch is a plug-in that greatly strengthens web search, X search, and single-page fetch. It works the moment it lands: the default engine is Firecrawl's keyless tier, 1,000 free credits every month, with no account, no API key, and no card.

Talk to us

Hit a problem? Open an issue. Everything else is welcome on X: @liustack. What you built with it, which harness you are on, what should come next. New releases land there first. A community space is on the way.

Features

• 🥇 The strongest free web search plugin for DeepSeek Harness (dsh): one command installs it, npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modsearch@5.10.11. In the desktop app's Add plugin dialog, enter @liustack/modsearch, not the GitHub URL. Details in harness setup.
• Free out of the box, no signup. Search and page fetch run on Firecrawl Keyless by default: 1,000 free credits/month, no account, no API key, no card. Every fallback channel is free too: Antigravity CLI needs only a browser sign-in, and Tavily, Exa, and a free Firecrawl key each add their own monthly quota with no card required.
• Automatic failover. When a channel fails or exhausts its quota, the next one takes over.
• Per-engine key rotation. Give Tavily, Exa, Firecrawl, or Keenable multiple comma-separated keys. Authentication, rate-limit, and quota failures rotate to the next key before the engine chain falls back.
• Searches X (Twitter). With Grok Build installed, ModSearch queries the corpus that web indexes cannot reach.
• Install once, use everywhere. Works in Claude Code, Codex, Pi, and OpenCode.

See it work

All four screenshots are unedited runs, the first two from the Codex desktop app and the last two from dsh web, driving DeepSeek models with no web access of their own.

Give it a blog link and ask what the post says. Twenty-five seconds later: a structured summary of the whole post, with no browser involved.

[Text-only DeepSeek summarising a blog link through ModSearch]

Give it no target at all, just "anything interesting in AI today?". Thirty-six seconds later: six sourced stories, with a closing note on which details came from aggregation and deserve a second look. The note comes from the uncertainty field.

[An open-ended question comes back as six sourced stories with a stated confidence caveat]

Ask dsh web for today's top AI stories. dsh's native search tool row runs straight on the modsearch engine chain, and eighteen seconds later three stories come back, each with a source link.

[dsh web's native search running on the modsearch engine chain, returning three sourced stories]

Ask which Node.js line is still in maintenance. read_page reads the release page and the release schedule in turn, and sixty seconds later the verdict arrives with a version status table and sources at the end.

[read_page reading two pages and returning the Node.js maintenance verdict]

Supported engines

Firecrawl, Keenable, and You.com work with zero setup. Every other engine is one command away. Keys live in ~/.modsearch/config.json (0600, masked when shown):

Keys can also come from the environment (TAVILY_API_KEY, EXA_API_KEY, FIRECRAWL_API_KEY, KEENABLE_API_KEY). One engine can take multiple keys as a comma-separated value such as key-one,key-two, in either the config file or its environment variable. Multiple engines configured means automatic failover, best first. Every engine participates by default. Exclude one with modsearch config set tavily.enabled false. Using a Tavily-, Exa-, Firecrawl-, or Keenable-compatible third-party or self-hosted endpoint? Point the engine at it: modsearch config set tavily.baseURL <url>. Official endpoints stay built into the code and are never written as default config. Every knob, engine by engine, is in the configuration guide.

Installation

Step 1, hand it to your AI. Search and page fetch work as soon as the skill lands, on Firecrawl's free keyless quota, so installation is one message:

Install and configure the modsearch skill following INSTALL.md at https://github.com/liustack/modsearch, then run the health check and tell me the result.

On DeepSeek Harness (dsh), send the same message. INSTALL.md recognizes dsh and switches to the native plugin instead of the skill, so dsh's own web_search runs on modsearch. In the browser UI your AI installs it. The desktop app only takes plugins from its own window, so there your AI asks you to open Plugins, click Add plugin, and enter @liustack/modsearch.

Prefer a direct command? Install the same skill through skills.sh:

npx skills add liustack/modsearch

Step 2 (optional), add more free engines. Antigravity CLI writes better synthesized answers. A free Tavily, Exa, or Firecrawl key adds a personal quota on top of the keyless one. None requires a card. agy's browser sign-in is the only step that needs your hands:

curl -fsSL https://antigravity.google/cli/install.sh | bash
agy                                                           # sign in, then exit

Picked a key instead? Send one line to your AI: "set my tavily key to tvly-...".

dsh users have a path that never touches the command line. Open Plugins in the dsh sidebar and pick @liustack/modsearch: pick the preferred engine, fill in an API key or a self-hosted endpoint, tick which engines join failover, hit save and it takes effect. On dsh releases before 0.1.6-alpha.2 the same card lives under Settings → Plugins → Plugin config.

[The ModSearch page in the dsh Plugins panel, shown in Chinese: pick the preferred engine, fill in an API key and endpoint, tick the engines that join failover]

Usage

Once installed, you do not need to remember any commands. Just chat. Ask anything that needs checking, or paste a URL, and the skill triggers on its own: it picks an engine, runs the search or fetch, and the answer comes back with sources.

Documentation

Proxy fake-IP mode (Clash, Clash Verge Rev, mihomo, Surge) works out of the box. DNS answers in 198.18.0.0/15 are treated as fake-IP placeholders with no switch needed. Literal URLs such as http://198.18.0.5/ stay blocked when allowPrivateNetwork is off. For other reserved ranges used by VPNs or hosts-file accelerators such as Steam++ / Watt Toolkit, see Blocked private network target.

Contributing

This repo does not accept pull requests. The project is maintained by the author alone. Every line is reviewed in person, and that review is the premise of its reliability. Two effective ways to contribute:

• Open an issue. Bugs, suggestions, confusing errors, unclear docs. Issues are read and shape what gets built next.
• Fork it. Under MIT your copy is fully yours to modify and publish.

Shameless plug

Follow the WeChat official account liustack (Chinese-language): AI tools, practice, and ideas, posted as they land. Scan the code, or search WeChat for "liustack":

⭐ If it helps, star ModSearch. Stars are how the next developer finds it.

Star History

Disclaimer

ModSearch is MIT-licensed, so use is not restricted. The author gives no warranty and no endorsement for any particular use, commercial or otherwise. The upstream engines it drives (Antigravity CLI, Tavily, Exa, Firecrawl, Keenable, You.com, Grok Build) each carry their own terms and quotas, and complying with them is the user's responsibility.

License

MIT