# scicomp-llm-mcp

Source: https://glimind.com/server/pypi%2Fscicomp-llm-mcp
Registry id: `pypi/scicomp-llm-mcp`
Status: unknown

## Install

```json
{
  "mcpServers": {
    "scicomp-llm-mcp": {
      "command": "uvx",
      "args": [
        "scicomp-llm-mcp"
      ]
    }
  }
}
```

Claude Code:

```sh
claude mcp add scicomp-llm-mcp -- uvx scicomp-llm-mcp
```

## Tools (32)

- `analyze_attention`: Analyze attention patterns
- `analyze_attention_patterns`: Detect attention patterns
- `analyze_data_influence`: Compute sample influence
- `analyze_norms`: Analyze layer norms
- `analyze_sequences`: Compute sequence statistics
- `analyze_sparsity`: Compute model sparsity
- `analyze_token_distribution`: Analyze token frequencies
- `analyze_weights`: Analyze weight distributions
- `compare_heads`: Compare attention heads
- `compare_models`: Compare model architectures
- `compute_gradient_norms`: Compute gradient norms
- `compute_head_rankings`: Rank heads by importance
- `compute_model_flops`: Compute model FLOPs
- `compute_perplexity`: Compute perplexity
- `create_model`: Create GPT or Mamba architecture
- `create_tokenizer`: Create or load tokenizer
- `create_trainer`: Configure training
- `estimate_memory`: Estimate training memory requirements
- `evaluate_model`: Evaluate on dataset
- `generate_text`: Generate text
- `get_model_config`: Get model configuration
- `get_training_status`: Monitor progress
- `list_models`: List all models
- `load_checkpoint`: Load from checkpoint
- `load_dataset`: Load training dataset
- `prepare_dataset`: Prepare for training
- `run_data_ablation`: Run ablation studies
- `save_checkpoint`: Save model checkpoint
- `suggest_augmentations`: Suggest data augmentation strategies
- `tokenize_text`: Tokenize text
- `train_step`: Execute training steps
- `visualize_attention`: Extract attention summary

Measured by Glimind (https://glimind.com): MCP handshake + tool list only, no tool is ever called. CC BY 4.0.
