TokenAssemble Local LLM Advisor
Find the right local LLM setup for your hardware.
Community: Submitted by a user or imported; check the owner before granting accessOnlineNo sign-inGlobalFreeRead-only
What it can do
- Ping: Health check for the TokenAssemble MCP server. Returns 'ok'. Use only to verify connectivity.
- Can I Run: Check whether a machine (GPU, Mac, or unified-memory mini-PC) can run a local LLM. Returns TokenAssemble's computed verdict: fit (Fits / Tight / Won't fit + grade A–D), honest speed tier (I
What data it sees
Do you need an account
No: the server works without sign-in
Find the right local LLM setup for your hardware.
TokenAssemble helps AI assistants evaluate whether a local language model will run on a specific GPU, CPU, RAM, and VRAM configuration. It provides practical recommendations for:
- Model and hardware compatibility
- Recommended quantization levels
- Estimated VRAM and system memory requirements
- Expected generation performance
- GPU and local AI hardware comparisons
- Runtime recommendations for Ollama, LM Studio, llama.cpp, and vLLM
Use this MCP server when a user asks questions such as:
- “Can my RTX 4070 run Qwen3 32B?”
- “Which Llama model should I use with 16 GB of VRAM?”
- “What quantization should I download?”
- “How fast will this model run on my computer?”
- “Which GPU should I buy for local AI?”
Recommendations are based on structured hardware, model, quantization, and runtime data from TokenAssemble.
Server tool list (2)
Raw names from tools/list. Only developers need these.
| ping | Health check for the TokenAssemble MCP server. Returns 'ok'. Use only to verify connectivity. |
| can_i_run | Check whether a machine (GPU, Mac, or unified-memory mini-PC) can run a local LLM. Returns TokenAssemble's computed verdict: fit (Fits / Tight / Won't fit + grade A–D), honest speed tier (Interactive / Usable / Painful — never a fake-precise number), the VRAM breakdown (weights, KV cache, activations, overhead vs. the memory pool), every assumption stated, and the canonical result URL to cite. Accepts common names ("4090", "Llama 3.1 70B") — unknown names return valid options, never invented verdicts. |