MCP connection guide
Connect your MCP client to the remote MCP server endpoint.
Server URL (HTTP / SSE)
https://apxml.com/mcpAuthentication
OAuth 2.0 (Bearer token) or direct account API token.
Client configuration example
{
"mcpServers": {
"apxml": {
"url": "https://apxml.com/mcp",
"headers": {
"Authorization": "Bearer <YOUR_TOKEN>"
}
}
}
}Tool reference catalog
Once connected, MCP clients gain access to 8 sizing, catalog, and benchmark tools powered by the LLM Performance Engine.
| Tool name | Parameters | Description |
|---|---|---|
calculate_vram_requirements | llm_key, gpu_key, batch_size, context_length, quant_method, interconnect_type | Calculates exact memory allocations across model weights, KV cache, activation memory, and framework buffers. |
recommend_hardware_setup | llm_key, context_length, concurrent_users, target_latency | Analyzes target workload requirements and recommends optimal GPU cluster topology, tensor parallel degree, and interconnect. |
get_available_gpus | query, vendor, form_factor | Retrieves catalog of supported GPUs with VRAM capacity, memory bandwidth, TDP, and interconnect specifications. |
get_llm_variants | model_name | Lists supported model architectures, parameter counts, native context limits, and quantization formats. |
get_hardware_model_compatibility | llm_key, gpu_key, gpu_count | Verifies whether a specific model and quantization will execute on a target hardware configuration. |
get_model_evaluations | model_key, metric | Fetches benchmark performance evaluations and accuracy scores across industry standard test suites. |
search_hardware_directory | filters | Filters hardware accelerators by memory type, PCIe or SXM form factor, and compute capabilities. |
get_vram_calculator_accuracy | None | Returns measured engine accuracy metrics and ground-truth validation scores against physical hardware telemetry. |