Get This Tool
Myco Brain
Pricing
- Model
- Free
Summary
Every agent session starts from scratch — the context you built, the decisions you logged, the facts you corrected, all gone the moment the conversation closes. Myco is a self-hosted memory layer that gives your agents a persistent, provenance-tracked store they can read and write across sessions and clients.
The core mechanic is deterministic writes: the application code writes facts to Myco's Postgres store, not the LLM, so every stored fact carries a source document, a confidence score, and a full audit trail queryable via brain_why. One MCP server exposes that memory to Claude Code, Cursor, Codex, Windsurf, and any other MCP-compatible client simultaneously — write from Claude Desktop, retrieve from Cursor, no sync step required. The vendor publishes a 500-question LongMemEval result and a recall@5 figure using a recency reranker, both on the full benchmark set. The hard ceiling appears when your agents need to act on what they remember — Myco stores and retrieves facts; it does not plan, route, or execute tasks, so orchestration logic lives elsewhere.
Bottom line: Pick Myco when you need auditable, shared memory across multiple MCP clients on your own infrastructure — plan a separate execution layer when your agents need to do something with what they remember.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Deterministic write path means the LLM never authors the facts stored in memory, so every retrieved fact links to a source document and confidence score — which means regulated teams get an audit trail without building one themselves.
- One MCP server shared across all connected clients, so a fact written from Claude Desktop is immediately readable by a Cursor agent without a sync job or intermediate API call.
- Full-stack boot with docker compose and no required API keys, so teams evaluate and prototype without committing credentials or cloud spend before the architecture is validated.
- Content-hash deduplication on document ingestion, so re-importing the same ChatGPT or Claude export twice does not corrupt or inflate the memory store.
- Graph queries over entity relationships via the built-in tools, so agents can retrieve not just isolated facts but the web of connections between people, decisions, and documents in the store.
Cons
Sign in to edit- Myco stores and retrieves facts — it has no planner, no task router, and no execution loop. Teams building agents that need to act on retrieved memory must implement that logic themselves, which means maintaining a separate orchestration layer alongside the memory layer.
- The self-hosted path requires running Postgres 16 with pgvector and managing that infrastructure. Teams without existing Postgres ops experience hit configuration and maintenance overhead that the single docker compose up does not eliminate long-term.
- Semantic search requires a local Ollama instance or an external embedding provider; teams without GPU-capable self-host infrastructure who want semantic recall beyond full-text search are blocked until the cloud offering exits beta — at which point they are evaluating a hosted product with a waitlist rather than a drop-in replacement.
- No API surface is exposed outside the MCP protocol, so teams whose agents run outside MCP-compatible clients cannot integrate without building a custom MCP wrapper — teams with that constraint typically move to a vector database with a standard REST or gRPC API instead.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- Postgres, Docker, MCP clients
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-06-20T03:32:44.924Z
Best For
Who it's for
- Developers building agents
- FinTech and regulated industries
- Users of Claude Code, Cursor, and MCP clients
What it does well
- Cross-session fact recall for agents
- Shared memory across multiple AI clients
- Audit trails for regulated decisions
- Graph queries over entity relationships
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Compare Myco Brain
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is Myco Brain free?
- Yes — Myco Brain is fully free to use. There is no paid tier.
- Is Myco Brain open source?
- Yes. Myco Brain is open source.
- Can I self-host Myco Brain?
- Yes. Myco Brain supports self-hosting on your own infrastructure.
- When was Myco Brain released?
- Myco Brain was first released in 2026.
- What platforms does Myco Brain support?
- Myco Brain is available on: Postgres, Docker, MCP clients.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
AI agents that rely on chat history for context break the moment a session ends, a new client connects, or a second agent joins the workflow. Myco addresses this by acting as a dedicated memory layer: it runs on your own Postgres 16 instance with pgvector, exposes a single MCP server, and lets any connected client read and write facts through three core tools — brain_search for retrieval, brain_context_pack for packing relevant context into a prompt window, and brain_why for tracing exactly which source documents support a given fact. The full stack boots with a single docker compose up and requires no API keys to start; local Ollama embeddings make semantic search keyless as well.
The differentiating design choice is that the program, not the LLM, writes facts to the store. This eliminates hallucinated memories at the write path — if a fact is in Myco, it traces to a source document, a confidence score, and an extraction record. The docs describe this as a ‘deterministic write path,’ and the brain_why tool surfaces the full provenance chain on demand. For FinTech teams or any regulated environment where you need to show a regulator why an agent said what it said, that audit trail is not a nice-to-have.
Myco fits cleanly into a stack where multiple agents or clients need to share facts — a decision logged by a Claude Desktop session is immediately retrievable by a Cursor agent without any manual sync. It also supports importing ChatGPT and Claude conversation exports, turning historical chat data into provenance-tracked memory with content-hash deduplication so re-imports do not multiply records. Where it does not fit: Myco is memory infrastructure, not an execution engine. Teams that need agents to branch logic, call external APIs, or chain tool calls based on retrieved facts need to wire that orchestration themselves — Myco has no planner, no task router, and no agent loop.
The self-hosted path is Apache-2.0 licensed and free. A cloud-hosted version is described on the vendor page as in beta waitlist status. Integration is documented for Claude Code, Claude Desktop, Cursor, Codex, Windsurf, Continue, Zed, and custom MCP agents — connection is a single command paste into the target client after the stack is running.