Myco: long-term memory for an AI agent
Myco
Agent memory that outlives model generations: an MCP operator, a Rust substrate and a Python kernel keep your history and context
- 67 ★ on the repository
- Last commit
Extended access
Why this level
- The agent writes material from your conversations and files into memory, where it stays as immutable history
- The system runs three runtimes on your machine and changes its own structure after your approval
Install
- Works in
Manual install
cargo build --release --workspaceBuild from the cloned repository, then follow docs/guides/GETTING_STARTED.md. Rust 1.80+, Node.js 22+ and Python 3.13+ are required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent writes material from your conversations and files into memory, where it stays as immutable history
- The system runs three runtimes on your machine and changes its own structure after your approval
This is third-party code. Review the repository files before installing.
What it does
Myco is designed as a memory layer that survives model swaps. You talk to the agent through MCP, and between turns the system runs one metabolic cycle: it takes new material as immutable history, consolidates understanding, prunes stale parts with a tombstone and proposes reshaping when the form no longer fits the work. The repository has three parts on one protocol. A TypeScript MCP operator the agent drives lives in operators/claude. A Rust substrate keeps history and the hash chain, and a Python kernel handles rules, governance and trajectory. The doctrine the system obeys is written in the repository and sealed with a hash chain so the rules cannot drift unnoticed. The system is keyless: trust rests on a live human at the gate and on a causal history that cannot be rewritten silently. The human approves any reshaping.
Who it is for. People who run long projects with an agent and want context, decisions and taste to persist across model changes.
Good fit when
- The agent forgets your decisions after a model change or a new window
- Months of notes have piled up and you need to see why a decision was made
- You want to experiment with long-term memory and a log that cannot be rewritten after the fact
- The project runs for years and context must live independent of the model version
Not a fit when
- You need stable ready-made memory for tomorrow workflow: the project is at v0.9 genesis alpha
- You need a one-command install: the build runs from source across three languages
- You need a managed cloud service with support: the project states it is not a managed service
Example request
Recall from the Myco log why we chose this database schema in April and note what has changed since.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Version v0.9 genesis alpha: the author says the system runs but awaits its first real cultivation. There is no one-command install: you need Rust 1.80 or newer, Node.js 22 and Python 3.13, clone the project and build it with cargo build, then follow GETTING_STARTED. Besides English, the documentation exists in Chinese and Japanese. The README does not list the tools of the MCP operator, details are in docs/guides/PILOT.md. The last commit dates from August 2026 and activity is low.
How to disable. Remove the Myco entry from the MCP client config and delete the cloned repository and build folders by hand. The accumulated history lives in project folders, so back it up before deleting.
FAQ
Do I need keys or signing to start?
No. The project says it is keyless: there is no owner key or signing ceremony, trust rests on a human at the gate and a causal history.
Which clients does it work with?
Claude Code through operators/claude, Claude Desktop and Cowork, and any MCP host over the standard protocol. The README names Cursor, Windsurf, Zed and OpenClaw.
Is it a vector database or a framework?
No, the README denies both, and a managed service too. Myco is described as a metabolism: it devours material, evolves, amplifies the agent and lets stale parts go.
What does the project consist of?
A TypeScript MCP operator, a Rust substrate and a Python kernel joined by one protocol. A shared test vector suite checks serialization parity across languages.
Who decides when to reshape the system?
The human. The system proposes a reshaping when its form no longer fits the work, and it applies only after your approval.
Discussion
No questions or reviews yet.
Related
MCP serversiai-memory
iai-memory (Personal Memory Engine)
A local long-term memory MCP server for agents: keeps conversations verbatim and returns the right context at the start of a session
A long-term memory MCP server for AI agents: facts fade along a forgetting curve and related records are compressed into a summary
A self-hosted MCP server for long-term agent memory: structured records, versioned rollback and a visual dashboard
An MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
A long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
Engineering team memory over MCP: decisions, bug causes and constraints recalled by every agent, from Claude Code to Cursor.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Myco: long-term memory for an AI agent (this card) | no | not required | 67 | |
| MCP Memory Keeper | no | not required | 136 | simple context storage between sessions without a hash chain |
| Agent Memory Engine | no | not required | 127 | a memory engine with retrieval, without a sealed doctrine or a causal history |
README in short
The manifesto describes the problem of the relationship with the agent restarting at every model change and proposes armor the agent inhabits. It includes an architecture diagram, a quick start, an integration list and guides for the human and for the piloting agent.