Myco: Agent-First Cognitive Substrate
Persistent memory layer for AI agents, with a validated knowledge graph and MCP access
The repository was last updated 98 days ago and is published under MIT.
- Pricing
- Open source
- Platforms
- CLI, self-hosted, API
- GitHub stars
- 68
- Last commit
- 3 months ago
- Language
- Rust
- License
- MIT
Similar MCP servers
About Myco: Agent-First Cognitive Substrate
Myco: Agent-First Cognitive Substrate is an open-source MCP server for knowledge and memory. It is available on CLI, self-hosted and API.
The README describes Myco as a persistent layer that an AI agent works inside across model changes. Between turns it ingests new material as immutable history, has the pilot turn that into durable understanding, and prunes stale parts with tombstones so outdated records do not accumulate. Changes to the overall structure go through an explicit approval step rather than happening silently.
The project is split across three parts in one repository: a TypeScript MCP operator that the agent drives, a Rust substrate that holds the data, and a Python kernel that runs the mechanisms. Its doctrine is stored in the same repo and sealed with a hash chain, and a self-checking layer looks for tampering, drift and runaway cost.
The README describes it as a v0.9 alpha that has not yet been used for real work. It is aimed at developers and individuals who want an agent to keep one person's context and decision history over a long period.
What it does
- Immutable history ingestion
- New material is stored as append-only history that the agent can later build on.
- Tombstone-based pruning
- Stale parts are removed with a tombstone record instead of being silently deleted.
- Approval-gated reshaping
- Structural changes to the graph wait for a human to approve them at the gate.
- Tamper-evident hash chain
- Causal history and the sealed doctrine are protected by a hash chain for tamper detection.
- Self-checking immune layer
- Built-in checks look for tampering, drift and excessive cost during operation.
- MCP integration
- Works with MCP hosts such as Claude Code, Claude Desktop and Cursor over the standard protocol.
Good for
- Keeping an agent's context about a long-running project across model upgrades
- Looking up why a past decision was made when records have drifted
- Maintaining a validated knowledge graph that an AI assistant reads and updates
- Testing agent memory behaviour in a Rust and Python codebase
Pricing
Open source
Free and MIT-licensed, self-hosted from the repository.
Myco: Agent-First Cognitive Substrate FAQ
Which MCP hosts does it support?
The README names Claude Code and Claude Desktop/Cowork directly, and says any MCP host such as Cursor, Windsurf, Zed and OpenClaw should work over the standard protocol.
Does it need an API key?
No. The README says the project is keyless, with no owner key or signing step to set up.
What languages make up the project?
The README describes a TypeScript MCP operator, a Rust substrate and a Python kernel, all in one repository.
What is its current status?
The README calls it a v0.9-genesis alpha that runs but has not yet had its first real use.
Is Myco: Agent-First Cognitive Substrate free?
Yes. Myco: Agent-First Cognitive Substrate is open source, so you can use and self-host it for free.
What are the best alternatives to Myco: Agent-First Cognitive Substrate?
The closest alternatives to Myco: Agent-First Cognitive Substrate are GrayMatter, Knowledge MCP, Compound Memory, codeagent-memory and mcp-memory-rs.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.