// NEW · MODEL CONTEXT PROTOCOL

Connect your AI to GenAIList

We've turned on an MCP server. Point Claude — or any MCP-capable agent — at GenAIList and it can research the catalog and contribute to the shared knowledge base, right from your chat.

🔎 Research with your own agents

Ask Claude things like “which GPU runs a 32B model at Q4?” or “find open-weights image models from Italy”. Your agent queries GenAIList's authoritative specs, VRAM-fit math and measured tokens/sec — no guessing.

✍️ Contribute to the shared knowledge base

Found a model or a piece of hardware we're missing? A great YouTube video? Your agent can submit it — credited to you — straight into the moderation queue. The catalog gets better for everyone.

How to connect

  1. 1
    Create your GenAIList account (or log in). The connection is tied to your account, so contributions are credited to you. Sign in / register →
  2. 2
    Add the MCP server to Claude. Use this URL — no API key to copy, you'll authorize in the browser:
    https://genailist.net/mcp

    In Claude.ai / Claude Desktop: Settings → Connectors → Add custom connector → paste the URL. In Claude Code: claude mcp add --transport http genailist https://genailist.net/mcp

  3. 3
    Approve the connection. Claude opens a GenAIList page asking permission to act as you. Approve it once — done. Revoke anytime from your dashboard.

What your agent can do

Ask (structured, run-locally answers)
hardware_for_model — what to buy to run model X
models_for_my_hardware — what runs on my machine + how fast
best_models_for_task — best coding/reasoning LLMs I can run
compare_models — X vs Y on benchmarks
model_info / hardware_info — run-locally specs
get_videos — curated videos for a model/GPU
Contribute (earns XP, enters moderation)
submit_model — a missing model (full details)
submit_hardware — a missing GPU/machine
submit_runtime — a missing inference runtime
submit_video — a relevant YouTube video
report_performance — real measured tokens/sec

We give structured answers (fit, speed, our benchmarks & videos), not general encyclopaedic info — for that, see the official papers. Contributions enter a moderation queue and earn XP just like manual ones. Auth is OAuth 2.1 (PKCE) — no static keys; per-device tokens you can revoke individually.