Myco: Persistent MCP substrate for agent memory and continuity
Myco, from Battam1111, is an MCP server that provides a persistent cognitive substrate to prevent context rot when agents switch models. It creates a self-evolving memory layer that ingests, indexes, and recalls agent state across sessions, preserving project understanding over time. The tool uses manifest-driven operations and a tamper-evident memory graph to manage continuity. It targets AI researchers, software developers, and power users needing long-running agent context.
What tasks can you actually use it for?
Myco serves as an agent-focused memory substrate that keeps conversational and project state alive across model changes and new sessions, addressing the problem of context loss. The design centers on agent workflows that need persistent state rather than single-shot prompts. The implementation exposes manifest-driven verbs to govern ingestion and lifecycle events; examples in the system include germinate, eat, assimilate, and sporulate to represent how inputs move into the substrate.
How reliable and auditable are its stored memories?
The tool stores state in a self-validating filesystem graph that provides a tamper-evident causal history, and it applies 25 lint dimensions that mechanically enforce contract invariants. This mechanical enforcement is described as a way to maintain predictable agent behavior and to surface violations automatically. The substrate is provider-agnostic, so the same memory layer persists when the underlying model generation changes, supporting continuity across different LLM hosts.
Does it fit into developer workflows without heavy changes?
Myco integrates with MCP-compatible clients such as Claude Desktop, Cursor, Windsurf, Zed, and VS Code, and it supports installation via npx, pip, or cloning and building with Cargo depending on the chosen runtime. Runtime requirements include Node.js 22+, Rust 1.80+, or Python 3.13+ for the corresponding install methods. The project is open-source and designed for extension, which aligns it with researcher and developer toolchains rather than end-user apps.
Good fit for technical teams building persistent agent projects
Myco is a focused choice for research and development teams that accept a developer-oriented setup and need cumulative agent memory across sessions. Plan for an integration phase and add explicit validation steps for the substrate’s evolving state before relying on it for critical decisions. A practical tip: define ingestion and verification procedures early and test continuity with MCP client workflows before production use.
