kaeru MCP
kaeru is an MCP server with an agent memory graph: ideas, outcomes and sources across sessions, local, with a team cloud option.
- 86 ★ on the repository
- Last commit
Extended access
Why this level
- The daemon stores graphs of thoughts and outcomes from your sessions and keeps a local server on `127.0.0.1:9876`.
- The enabled cloud tier sends selected nodes to a trusted group.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/LamantinAI/kaeru/main/contrib/install/install.sh | bashThe first install command from the README installs the binary and service on Linux and macOS. On Windows download the archive from the releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The daemon stores graphs of thoughts and outcomes from your sessions and keeps a local server on `127.0.0.1:9876`.
- The enabled cloud tier sends selected nodes to a trusted group.
This is third-party code. Review the repository files before installing.
What it does
kaeru is a memory engine for agents in Rust: a typed graph where an agent thinks plus a recollection layer for long-term ideas and outcomes. Data lives locally in RocksDB and by default nothing leaves the machine. One `kaeru-mcp` daemon owns the vault, and sessions from Claude Code, Opencode, Cursor and others connect to it over HTTP or through `kaeru-mcp --stdio` as a relay. The shared API has 35 named operations (`context`, `search`, `read`, `note`, `claim`, `synthesise`, `history`, `lifecycle` and more): 16 frequent, ten local maintenance and nine optional cloud ones. Memory splits into operational and archival tiers, nodes keep assertion and retraction history, and projects are separated by initiatives. The optional `kaeru-cloud` gives a team shared storage: a node goes there only under an explicit policy and after a scan for keys and tokens. The interface at `http://127.0.0.1:9876/ui/` renders the knowledge graph.
Who it is for. Teams and developers running several agents on shared projects who want memory and reasoning history to survive sessions.
Good fit when
- An agent opens a project and must recall what it was thinking last time and follow provenance chains.
- Experiment outcomes are promoted from working memory into stable knowledge.
- Several agents and people on one project share nodes through a team cloud.
- Before asking the user, an agent checks memory through the `kaeru-first` hook.
Not a fit when
- You need a plain memory file with no separate daemon: kaeru runs a process and a RocksDB vault.
- You plan to host kaeru as a service for third parties: the BSL 1.1 license forbids it.
- A device on Intel Mac or Linux ARM: there are no prebuilt binaries, so build from source.
Example request
Connect to kaeru, read the project context for the `billing` initiative, find earlier decisions about payment retries and record the outcome of today experiment.MCP
- Transport
- http, stdio
- Authentication
- not required
Environment variables
KAERU_VAULT_PATH- Vault path instead of the default directory.
KAERU_MCP_UPDATE_CHECK- The value `0` disables the daily update check.
Limitations
Business Source License 1.1: the code is open, you may run it and embed it in your own products, but you may not hand kaeru itself to third parties as a managed service; each version converts to Apache 2.0 after four years. The `curl | bash` installer downloads a release, installs `kaeru-mcp` into `~/.local/bin` and creates a user service. The daemon checks for releases daily and sends the version, platform and install channel, never vault content, and `KAERU_MCP_UPDATE_CHECK=0` disables the check. The cloud tier is for a trusted group for now, without per-user isolation.
How to disable. Stop the service (`systemctl --user stop kaeru-mcp` on Linux or `launchctl` on macOS), remove the connection with `claude mcp remove kaeru` and delete the binary from `~/.local/bin`. Delete the `KAERU_VAULT_PATH` vault separately if you no longer need it.
FAQ
How do I connect kaeru to Claude Code?
Run `claude mcp add --transport http kaeru http://127.0.0.1:9876/mcp` with the daemon running. For stdio clients there is `kaeru-mcp --stdio`.
Why one daemon per vault?
RocksDB is single-writer, so the daemon owns the vault and every agent session connects to it without opening a second copy.
Does memory go to the cloud on its own?
No. The default initiative policy is `private`, and a node goes to the cloud only after an explicit mark and a scan for keys, tokens and private keys.
Which agents are supported?
Claude Code, Opencode, Cursor and any MCP client. Opencode has `contrib/opencode/install-opencode.sh`, and others can use the rules from `skills/kaeru-skill/SKILL.md`.
How do I turn off update checks?
Set `KAERU_MCP_UPDATE_CHECK=0`. The daemon stops sending the version and platform.
Discussion
No questions or reviews yet.
Related
MCP serversA long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
Shared memory for fleets of AI agents with access scopes, an audit trail and a knowledge graph: an MCP server, Docker and a cloud
Local memory for agents in a single Rust binary: stores, searches by meaning and keyword, works with no network and no keys
A local MCP server with five verbs that keeps project decisions across sessions: store, retrieve, supersede and audit diffs.
Local memory for agents: decisions from browser chats reach Claude Code, Cursor and Windsurf through MCP
An MCP server and API of shared memory for several agents: facts, decisions and events live in one database with semantic search.
README in short
The README describes the two memory tiers, lists features and API operations, gives installation, agent connection, local and cloud modes, the visualizer and the license.