bettermemory
<p align="center">
<img src="https://raw.githubusercontent.com/0Mattias/bettermemory/main/docs/assets/banner.svg" alt="bettermemory: memory that is checked before it is believed" width="100%">
</p>
<p align="center">
<a href="https://pypi.org/project/bettermemory/"><img src="https://img.shields.io/pypi/v/bettermemory.svg" alt="PyPI"></a>
<a href="https://github.com/0Mattias/bettermemory/actions/workflows/ci.yml"><img src="https://github.com/0Mattias/bettermemory/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
<a href="https://www.python.org/downloads/"><img src="https://img.shields.io/badge/python-3.11%E2%80%933.14-blue.svg" alt="Python 3.11-3.14"></a>
<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT"></a>
<a href="https://glama.ai/mcp/servers/0Mattias/bettermemory"><img src="https://glama.ai/mcp/servers/0Mattias/bettermemory/badges/score.svg" alt="bettermemory MCP server"></a>
</p>
<!-- mcp-name: io.github.0Mattias/bettermemory -->
An MCP memory server for coding agents. Stored facts get checked against
the filesystem and git *before* the model relies on them, so a memory
that has rotted is flagged instead of quoted back at you.
## Install
Claude Code — two slash commands, zero config:
```
/plugin marketplace add 0Mattias/bettermemory
/plugin install bettermemory@bettermemory
```
Anything else that speaks MCP:
```sh
uv tool install bettermemory
bettermemory init --client claude-desktop # or cursor / cline / claude-code / hermes
```
Restart the client. `bettermemory doctor` exits 0 when it's wired
correctly and prints a one-line fix for anything that isn't.
Try it without installing: `uvx bettermemory try` writes a memory
citing a file, deletes the file, and shows the next search flagging it.
Offline, throwaway store.
## What it does
- Checks memory before believing it. Every hit carries a staleness
verdict: calendar age, whether the paths it cites still exist, the
commits landed since it was last confirmed, and how the record
entered the store (written here, pulled, or placed by hand).
Declared claims
(`path`, `path::symbol`, `path::NAME=literal`) are re-checked
against the working tree; a claim that stops being true blocks the
stamp instead of riding along.
- Retrieval is deliberate. Memory is a tool call, not an
injection. Write gates bounce transient state, secret-shaped tokens
and near-duplicates; claims about *you* stage for confirmation.
- The code is the model. Search is deterministic lexical ranking —
keyword + BM25, fused — over your own vocabulary. No embedding
models, no downloads, nothing to warm up, same answer every time.
- Plain files. One markdown file per memory. Greppable,
git-syncable, no cloud, no account. The SQLite index beside the
files is a derived cache you can delete; `bettermemory reindex`
rebuilds it.
- Rot gets acted on, not accumulated. Episodes journal per-session
run-state without polluting durable search; health telemetry and
curation tools surface what drifted, what went cold, and what
contradicts what.
- Receipts, not adjectives. The claims above are measured by
preregistered benchmarks whose frozen result artifacts — misses
included — live in [bench/][bench] (`results/*.json`, each carrying
the declared terms it was scored under), and `bettermemory eval`
scores whether memory actually helped against your own log ([ours
is published][eval-results]).
## For agents
You were likely handed this link with "set this up." On Claude Code,
the two plugin commands above are the whole install — skip to step 3.
Everywhere else, steps are idempotent and safe to re-run:
1. `uv tool install bettermemory` — or `pipx install bettermemory`,
or `pip install bettermemory` into a venv. Python 3.11–3.14.
2. `bettermemory init --client claude-code` — likewise
`claude-desktop`, `cursor`, `cline`, `hermes`. For any other MCP host, plain
`bettermemory init` prints the canonical `mcpServers` JSON snippet
plus known config locations. (Continue needs a hand-written YAML
entry — see [docs/clients.md][clients-continue].)
3. `bettermemory doctor` — exit 0 means correctly wired; every failed
check prints a one-line fix.
4. Have the user restart the client so the server loads, then confirm
by asking the model *"what memory tools do you have?"*
Your operating contract — tool signatures, retrieval discipline,
write gates — is [docs/api.md][api]; the server's
`instructions` block delivers the core policy automatically. For the
long-form policy in your system prompt: `bettermemory init
--with-addendum`. Migrating from Claude Code's built-in auto-memory:
`bettermemory ingest` imports those files once.
## Everything else
The rest is written down: mechanics, storage format, the full tool
surface, CLI, configuration and limitations in
[internals][internals]; per-client setup in [clients][clients] and
[installation][installation]; eval methodology in [eval][eval];
postmortems in [incidents][incidents]; release history in
[CHANGELOG][changelog]; dev setup and the compatibility contract in
[CONTRIBUTING][contributing].
MIT licensed — see [LICENSE][license].
[api]: https://github.com/0Mattias/bettermemory/blob/main/docs/api.md
[bench]: https://github.com/0Mattias/bettermemory/tree/main/bench
[changelog]: https://github.com/0Mattias/bettermemory/blob/main/CHANGELOG.md
[clients]: https://github.com/0Mattias/bettermemory/blob/main/docs/clients.md
[clients-continue]: https://github.com/0Mattias/bettermemory/blob/main/docs/clients.md#continue-legacy-shape--see-caveat
[contributing]: https://github.com/0Mattias/bettermemory/blob/main/CONTRIBUTING.md
[eval]: https://github.com/0Mattias/bettermemory/blob/main/docs/eval.md
[eval-results]: https://github.com/0Mattias/bettermemory/blob/main/docs/eval-results.md
[incidents]: https://github.com/0Mattias/bettermemory/blob/main/docs/incidents/
[installation]: https://github.com/0Mattias/bettermemory/blob/main/docs/installation.md
[internals]: https://github.com/0Mattias/bettermemory/blob/main/docs/internals.md
[license]: https://github.com/0Mattias/bettermemory/blob/main/LICENSE
TDQS
Scored across 18 tools
Each tool targets a clearly distinct resource and action (search vs list vs show; write vs update vs remove vs verify; memory vs episode tier; scope enable/disable). Memory_write_confirm and memory_write_cancel are separate and unambiguous, and memory_audit_turn is explicitly not for model use.
All tools follow a consistent pattern: memory_<verb> and episode_<verb>, using snake_case throughout. Verbs are descriptive (write, update, remove, verify, search, list, show). No mixed conventions or vague verbs.
18 tools is on the heavier side but appropriate for a memory server covering memory lifecycle, scoping, verification, usage tracking, and an episode tier. Each tool has a distinct role; no redundancy. Slightly above the ideal 3-15 range but justified by the breadth of the domain.
Core CRUD and verification flows are fully covered (write, confirm/cancel, update, remove, verify, restore via reference). However, tools referenced in descriptions like memory_restore, memory_list_tombstones, memory_conflicts, and memory_proposals are not actually exposed, creating minor gaps. Agents can work around these by using memory_remove and memory_write, but the surface is not entirely self-contained.