Home / MCP server

RigBase MCP server

Last updated: 1 Nov 2026

Free, public, read-only Model Context Protocol endpoint so AI assistants (Claude, Cursor, ChatGPT, Gemini / Google Connected Apps, and others) can query 187 measured local-LLM builds and 442 tok/s results — and always link back to RigBase.

Tools

1. search_builds — filter measured builds by GPU, model, min VRAM, max price; returns tok/s, quant, engine, source credit and canonical URL.
2. can_it_run — given GPU(s) + count and a model, return measured results if any; otherwise a VRAM-fit estimate (labelled as estimate). Advises against quants below Q4.
3. best_gpu_for_model — fastest measured result for a model, plus tok/s per euro where a build cost or GPU street price is known.
4. get_build — one build by slug, with all measured results, credit and URL.

Every tool result includes the original source credit (e.g. Mia's AI Lab, Tech2wild) and a https://getrigbase.com/… link. No auth. Read-only. No user data collected.

Example prompts

• “Using RigBase, what is the fastest measured setup for Qwen3.8 27B?”
• “Can 4× RTX 3090 run gpt-oss 120B? Check RigBase measured builds first.”
• “Search RigBase for DGX Spark builds of DeepSeek V4.1 Flash and cite the sources.”
• “Best GPU for Gemma 4 31B on RigBase, with tok/s per euro if known.”

Add it to a client

Cursor

In Cursor MCP settings, add a remote server URL:

https://getrigbase.com/mcp

Or in ~/.cursor/mcp.json / project .cursor/mcp.json:

{
  "mcpServers": {
    "rigbase": {
      "url": "https://getrigbase.com/mcp"
    }
  }
}

Claude (Desktop / Claude.ai custom connectors)

Add a custom MCP connector with URL https://getrigbase.com/mcp (Streamable HTTP). No API key.

ChatGPT (Developer mode / MCP)

Where custom MCP servers are supported, use the same URL. Tools are read-only.

Gemini / Google AI Mode (Connected Apps)

Google's Connected Apps Platform accepts remote MCP servers over public HTTPS. RigBase is preparing a Connected App listing (domain verification via /.well-known/google-connected-apps needs Google Connected Console access, currently waitlisted outside the US). Meanwhile you can point Gemini Developer Mode at https://getrigbase.com/mcp.

Try it with curl

curl -sS https://getrigbase.com/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

curl -sS https://getrigbase.com/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

curl -sS https://getrigbase.com/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"best_gpu_for_model","arguments":{"model":"Qwen3.8 27B"}}}'

About the data

Numbers are measured tok/s copied from public sources (benchmarks, posts, open recipes). RigBase does not invent or extrapolate measured speeds. VRAM “can it run” answers without a matching build are arithmetic estimates and are labelled as such. Always credit the original source named in the tool result and link the RigBase URL.

Human site: getrigbase.com · Free dataset: /data/ · Contact: [email protected]