Your model doesn't know about this week.
It can't tell you about a framework release or a doc change that landed after its training cutoff. Grounder hands it current, cited Google answers, so it can.
Your local model is frozen at its training cutoff. Grounder gives it Google's live answers, the real AI Overview and AI Mode, with citations, as three MCP tools in the stack you already run.
You went local to keep your work yours. Grounder never stores what you searched.
Private alpha. No subscription lock-in, no contact-sales funnel.
Grounder is a privacy-first Google search layer for local LLMs and MCP agents.
Your model calls it when it needs something current. Grounder returns Google's organic results, AI Overview, and AI Mode as three MCP tools, with the citations to back every answer.
No cloud-search bargain: no hidden retention, no telemetry, no sprawling tool catalog your agent has to fight. Just Google's current answer, in the local stack you already run.
It can't tell you about a framework release or a doc change that landed after its training cutoff. Grounder hands it current, cited Google answers, so it can.
Three MCP tools your agent calls directly, instead of a 70-engine catalog it has to prompt its way around.
Grounder fetches your query and drops it. Nothing about what you asked is kept, logged, or used to profile you.
Intentionally narrow. Your model shouldn't have to pick between dozens of overlapping search tools.
google_searchGoogle's organic results and the context around them: titles, snippets, source URLs, related and People Also Ask questions. For when the agent wants source diversity or links to read further.
google_ai_overviewGoogle's AI Overview when it shows one, as clean text with its citations. The concise answer Google places above the results.
google_ai_modeGoogle's AI Mode: a deeper, multi-source answer with a larger citation set. For more breadth than an AI Overview.
You already run a local stack (Ollama, LM Studio, Continue.dev, Aider) and want current Google answers without routing your work through a cloud middleman.
You want a broad SERP-scraping platform for SEO rank-tracking, or a full cloud agent stack. Grounder is a focused search layer, deliberately neither.
Grounder processes your query to fetch Google results, then drops it.
We don't keep the text of your searches for analytics, training, profiling, or telemetry. The only records we hold are billing counters and abuse-prevention events, and neither contains a word of what you searched.
| Need | Grounder | Cloud search APIs |
|---|---|---|
| Local LLM fit | Built around MCP and local clients. | Built around cloud SDKs, dashboards, and broad catalogs. |
| Answer source | Google's AI Overview, AI Mode, and organic results, verbatim. | Their own crawl, or a synthesis their model wrote. |
| Your queries | Fetched, then dropped. Billing records only. | Often retained, with telemetry on by default. |
| Tool surface | Three tools your model can reason about. | Large surfaces agents fumble or over-prompt. |
| Pricing | One call, one credit. Packs, no sales gate. | Tiers, bundles, multipliers, "contact sales." |
Add Grounder to any MCP client with your API key. One block, no SDK.
{
"mcpServers": {
"Grounder": {
"command": "uvx",
"args": ["grounder-mcp"],
"env": { "GROUNDER_API_KEY": "your_key_here" }
}
}
}
A privacy-first Google AI search layer for local LLMs and MCP agents. It gives local models current Google AI Overview, AI Mode, and organic search results with citations.
No. Grounder doesn't host your model. It provides search and citation context that your existing local model or MCP-compatible agent can call.
No. Grounder processes query content to fetch Google results, but does not store it for telemetry, analytics, training, or profiling.
Yes, if your priority is local-LLM fit, Google's actual AI Overview and AI Mode (not a synthesis), and no stored query content. Tavily and friends crawl their own index and can store your queries.
Any MCP-compatible client, including local AI tools such as Ollama, LM Studio, Continue.dev, Aider, Claude Desktop, and Cursor.