{"repo":"norika1207-lab/mercury-mcp","free":true,"listed":false,"github":"https://github.com/norika1207-lab/mercury-mcp","clone":"git clone https://github.com/norika1207-lab/mercury-mcp.git","description":"Cross-architecture LLM internal observation database (23 models, 13 architecture families). Exposed as MCP tools for any AI coding agent.","language":"Python","stars":43,"topics":["ai-agents","consumer-hardware","cross-architecture","llm","mcp","mechanistic-interpretability","model-context-protocol","open-data","transformer","anchor-dimensions"],"license":null,"category":"mcp-servers","readme_excerpt":"Mercury MCP: Cross-Architecture LLM Internal Observation, as Agent Tools \"Most AI coding agents don't know what's inside the model they're talking to. Mercury does.\" Mercury MCP exposes a 23-LLM cross-architecture observation database to any agent that speaks the Model Context Protocol (Claude Code, Cursor, Cline, Goose, etc.). Built entirely on consumer hardware (one Mac mini + one NVIDIA DGX Spark) at near-zero compute cost. --- What it answers Try these prompts with any MCP-aware agent after installing: - \"What hidden dimensions are universally hot across LLM families?\" - \"In qwen-7B, which layer has the same functional fingerprint as falcon-7B layer 16?\" - \"Compose a layer recipe for a Chinese-writing + reasoning hybrid model.\" - \"Show me OLMo2's anchor dimensions and how they overlap with the qwen anchor set.\" The agent calls Mercury MCP tools; Mercury answers from precomputed observation data. --- Why this matters For mech interp researchers 23-LLM cross-architecture survey across 13 architecture families, two observation tiers per model. Tier-A (output-layer logit hooks): cheap, fast, candidate-signal screening. May surface artifacts from shared tokenization (vocab-id mod hidden size collisions). Useful as hypothesis-generating layer. Tier-B (HF output hidden states across all layers): the actual finding layer. Per-layer residual stream fingerprints. Cross-architecture functional layer alignment qwen-7B L15 to falcon-7B L16 reaches 0.868 similarity. 54/84 model-pairs a","default_branch":null,"files":null,"tree":[],"storefront":"/r/norika1207-lab","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/norika1207-lab/mercury-mcp/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}