attention-matters
Geometric memory engine on the S³ hypersphere — quaternion drift, phasor interference, MCP server
claude mcp add --transport stdio srobinson-attention-matters npx -y attention-matters serve
How to use
attention-matters provides a Model Context Protocol (MCP) server that exposes a persistent geometric memory workspace. Built as a Rust-based implementation with a CLI and server tooling, you interact with it through the provided CLI commands (am_query, am_buffer, am_ingest, am_salient, am_feedback, am_activate_response, am_batch_query, am_export, am_import, am_stats) to recall, ingest, inspect, and export memory state. The server surfaces a geometric memory engine where words live on the 4D S^3 manifold, and recall emerges from drift, interference, and phase coupling across conscious and subconscious manifolds. Tools like am_query and am_batch_query let you query the memory surface, while am_ingest and am_buffer manage content ingestion and preparation for recall. The export/import commands enable portable state management, and am_stats/inspect provide visibility into the current geometry and system health.
How to install
Prerequisites:
- Node.js and npm (for npx usage of the MCP server) or Rust toolchain (for building from source).
- Optional: cargo installed (Rust), if you plan to build from source.
Install and run using prebuilt MCP server via npm/npx:
- Ensure Node.js and npm are installed.
- Run:
npx -y attention-matters serve
If you prefer to build from source (Rust crates):
- Install Rust toolchain (https://www.rust-lang.org/tools/install).
- Build and install the CLI server:
cargo install --path crates/am-cli
- Run the server (via the installed am CLI or your preferred invocation):
am serve
Prerequisites summary:
- Node.js + npm for npx-based usage
- Rust toolchain + cargo for building from source
- Optional: familiarity with am CLI tools (am_query, am_buffer, am_ingest, etc.)
Additional notes
Tips and notes:
- The server is designed to preserve and surface geometric memory across sessions; use am_export to capture a portable state and am_import to restore.
- You can inspect the current geometry with am inspect neighborhoods and monitor system state with am stats.
- The architecture emphasizes two conservation laws: M = 1 (total mass) and K_CON + K_SUB = 1 (zero-sum coupling between manifolds). Adjusting these constants affects how the memory surface evolves over time.
- When importing state from the original implementation, compatibility is maintained with the v0.7.2 wire format, so legacy state files can be loaded.
- If you run into issues with npx caching or package resolution, consider clearing npm cache or installing directly via cargo as per the build-from-source steps above.
Related MCP Servers
mcp-agent
Build effective agents using Model Context Protocol and simple workflow patterns
openapi
OpenAPI definitions, converters and LLM function calling schema composer.
neurolink
Universal AI Development Platform with MCP server integration, multi-provider support, and professional CLI. Build, test, and deploy AI applications with multiple ai providers.
omega-memory
Persistent memory for AI coding agents
goai
AI SDK for building AI-powered applications in Go
elasticsearch-memory
🧠 Elasticsearch-powered MCP server with hierarchical memory categorization, intelligent auto-detection, and batch review capabilities