MentisDB MCP
Durable memory for agents: an MCP daemon with a hash-chained thought log and a shared versioned skill registry.
- 91 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The daemon stores everything agents write into it and requires no authentication by default.
- Open ports and shared memory across several agents widen who can read what was written.
Install
- Works in
Manual install
cargo install mentisdb --features local-embeddingsRust 1.90 or newer is needed; without the flag text hashing replaces semantic embeddings.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The daemon stores everything agents write into it and requires no authentication by default.
- Open ports and shared memory across several agents widen who can read what was written.
This is third-party code. Review the repository files before installing.
What it does
MentisDB is a memory daemon and versioned skill registry for AI agents. Agents write typed thoughts with hashes, tags, concepts and importance into an immutable log, then search them by meaning, agent identity, role and time. Several agents and tools connect to one chain and share memory. When a context window fills, an agent writes a summary, compacts history and reloads `mentisdb_standing_context` and `mentisdb_recent_context`. The skill registry stores each upload as a new version as a diff and lets you reconstruct any older one. Ed25519 signatures verify a version's author. The daemon exposes 45 MCP tools, REST, a web dashboard and a terminal UI.
Who it is for. Developers running long projects in Claude Code, Codex or Cursor who want shared memory across sessions, tools and several agents.
Good fit when
- An agent's context overflowed and project decisions and traps must not be lost.
- Several agents work on one repository and must read and append to a shared memory.
- A team wants versioned agent instructions with history and author signatures.
- You need a verifiable decision history: a hash-chained log rules out silent rewrites.
Not a fit when
- A short notes file in the project is enough: the daemon is overkill.
- A local service with open ports on the machine is not acceptable.
- You need memory on cloud storage without your own server.
Example request
Load recent context from MentisDB, tell me where we left off and write the result of the current step into memory.MCP
- Transport
- stdio, http
- Authentication
- not required
Environment variables
MENTISDB_BEARER_TOKEN_ACCESS- Turns on token checks for MCP and REST; without it access is open to anyone who reaches the port.
MENTISDB_DASHBOARD_PINsecret- PIN for the web dashboard; set it before exposing it outside.
MENTISDB_DIR- Data directory: chains, skill registry, certificates and the model cache.
Limitations
A fresh daemon has no authentication. In Docker it listens on `0.0.0.0`, so ports are published on loopback only, for example `-p 127.0.0.1:9471:9471`. For access from outside, set `MENTISDB_BEARER_TOKEN_ACCESS=true`, create tokens and set `MENTISDB_DASHBOARD_PIN`. Building from source needs Rust 1.90 or newer and system packages, on Ubuntu libssl-dev and libasound2-dev. The full build with local embeddings needs glibc 2.39; for older systems use the compat builds or Docker. The skill registry is integrity-checked as a file and is not hash-chained, unlike thoughts. Last commit 11 November 2026.
How to disable. Remove the MCP entry from the client, for example `claude mcp remove mentisdb`, and stop the daemon. The `MENTISDB_DIR` data directory or Docker volume holds chains and the registry; clear them separately.
FAQ
How do I install MentisDB?
Run `cargo install mentisdb --features local-embeddings`, then `mentisdb wizard` connects the tools it finds. Prebuilt release binaries and the Docker image `ghcr.io/cloudllm-ai/mentisdb:latest` exist.
How do I connect MentisDB to Claude Code?
Start the daemon and run `claude mcp add --transport http mentisdb http://127.0.0.1:9471`. Automatic setup: `mentisdb setup claude-code`.
How do I run it in Docker safely?
Publish ports on loopback only, as in `-p 127.0.0.1:9471:9471`. Otherwise an unauthenticated store is exposed to any machine that can reach the host.
What is mentisdb_recent_context?
An MCP tool that returns the chain's recent thoughts. The agent calls it after compacting context to recall where it stopped.
Does Claude Desktop need a daemon?
No. `mentisdb setup claude-desktop` writes a stdio config, and the process finds a running daemon or starts one in the background. Node.js and mcp-remote are not needed.
Discussion
No questions or reviews yet.
Related
MCP serversMenteDB keeps agent memory in a Rust database: an MCP server for Claude Code and Cursor in cloud or self-hosted mode.
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
MCP server with a memory of the agent's own: it decides what to keep and wakes with it each session, stored in one SQLite file
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Local memory for agents with a review queue: SQLite, an audit log and a token ledger, one vault for Claude Code, Codex and Cursor
A long-term memory MCP server for AI agents: facts fade along a forgetting curve and related records are compressed into a summary
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MentisDB (this card) | no | not required | 91 | |
| Basic Memory | no | not required | 4,566 | keeps notes in Markdown files; MentisDB writes typed thoughts into a hash-chained log and holds a skill version registry |
| MCP Memory Service | no | not required | 2,117 | serves memory for several clients; MentisDB adds an immutable chain, Ed25519 signatures and versioned skills |
README in short
The document explains the project's ideas, install paths through cargo, binaries and Docker, quick client setup and daemon configuration. Separate sections cover search semantics, the web dashboard, TLS, backups, the MCP tool catalog and the skill registry.