Grounder MCP (io.github.rozetyp/grounder)
Grounder MCP provides live web grounding for local and cloud LLMs via four MCP tools. It supplies ranked web search results, real page content, and a cited evidence pack that is token-capped to fit a model’s context window. The service uses the current web while noting that each model is frozen at its training cutoff.
🛠️ Key Features
- Four MCP tools for web grounding
- Ranked results and real page content
- Cited evidence packs sized to a context window
- Token-capped evidence output
- Thin stdio client; hosted service at https://grounder.dev
🚀 Use Cases
- Retrieval-augmented generation (RAG) with live web evidence
- Grounding answers for local LLMs or cloud LLMs
- Research workflows requiring cited web content
⚡ Developer Benefits
- No browser and no scraper; nothing heavy runs locally
- Works as an MCP server/client via stdio
- Shared demo key with no signup for demo use
⚠️ Limitations
- Requires access to the hosted service at https://grounder.dev
- Uses a shared demo key by default; real use needs a free key (1,500 pages/month, no card)
- Models remain limited to their training cutoff; Grounder supplies current web content as evidence
Topics
ai-agents, claude, cursor, grounding, llm, lm-studio, local-llm, mcp, mcp-server, model-context-protocol, ollama, rag, retrieval, search-api, web-search