NeuronScope

CLI and MCP server for mechanistic interpretability: trace which attention heads and neurons drive an LLM's output, built on TransformerLens. Schema-versioned JSON output, agent-callable via MCP, activation patching, zero-ablation, circuit sketches for GPT-2, Llama, Gemma, Qwen, Pythia and 240+ open-weight models.

{
  "mcpServers": {
    "neuronscope": {
      "command": "uvx",
      "args": [
        "neuronscope"
      ]
    }
  }
}

claude mcp add neuronscope uvx neuronscope

{
  "mcpServers": {
    "neuronscope": {
      "command": "uvx",
      "args": [
        "neuronscope"
      ]
    }
  }
}