Monet MCP Server

Monet · local-first memory for coding agents

Your coding agent shouldn't wake up blank.

Monet gives it durable, local-first memory — so a generic agent learns how you work, stops re-asking, and starts every session smarter than the last. 100% on your machine.

Paste this into Claude Code, Cursor, or any MCP agent:

Set up Monet for all my projects: read https://raw.githubusercontent.com/team-monet/with-monet/main/bootstrap/install.md and follow it, checking with me at each step.Copy

Star with-monet Preview the dashboard

The problem

Context windows are not memory.

Every new session, your agent forgets your codebase, re-asks what you settled yesterday, and drifts from decisions you already made.You pay the tax in re-explaining — over and over.

See it in action

This is a warm start.

Watch an agent pick up your project mid-stream — restoring context, even tidying its own memory, before it answers.

agent session

worked for 5m 19sReplay

you can we start working on monet? what could be next item to focus on?

agent I'll restore the Monet context first so I'm not guessing from a cold start, then suggest a next focus from the active workstreams and open threads.

agent_context

agent The restored thread says the live code work is basically done; the open edge is mostly propagation and dogfooding. Pulling the connected context so the pick isn't just “do the next checkbox.”

memory_gather

agent The connected memories cluster into three: propagation and dogfood, onboarding friction, and store hygiene. Reading the dogfood and onboarding notes before ranking them.

memory_fetch

agent One housekeeping thing surfaced while reading — a concept had new evidence attached but not yet synthesized. Folding it into the living memory now, so future sessions don't have to reconcile stale notes.

memory_synthesize

agent Spotted a stale line I'd just carried forward — correcting it before I hand you a recommendation.

agent Saving where we are so the next session picks up exactly here.

memory_checkpoint

agent

The next best item is a dogfood pass, using the cleanup backlog as the real workload.

The core mechanics are basically done — the open question is whether the recent change actually prevents missed memory under load, and the backlog is a concrete task that tests exactly that.

I'd triage the hygiene backlog, use it as live dogfood, then decide whether the prompt-only change is enough. It pays down real mess — and tells us whether Monet is teaching its own operator to remember.

An actual Monet session — sensitive details trimmed.

What Monet does

Living project state, not a pile of transcripts.

Monet turns scattered observations into structured state your agent can gather, fetch, and update — decisions, conventions, errors, and how they connect.

Local-first

One SQLite store at ~/.monet. On your machine, organized per project. No cloud.

MCP-native

Speaks MCP, so it plugs into the agents you already use.

Observation → concept

Append-only evidence rolls up into living concepts that stay current.

Graph gather

Pulls connected context across a work thread, not just keyword-matched snippets.

with-monet · the harness

Memory is only half of it.

Raw memory tools don't help if your agent doesn't know when to use them. with-monet installs Monet in one line and brings a memory-aware agent team that runs the loop for you — pulling the right context into every task, capturing decisions and gotchas as they happen, and never re-asking what you've already settled. The playbook handles setup for you.

This is what turns a generic coding agent into yours.

Star with-monet on GitHub

Local-first · verify it yourself

Your code never leaves your machine.

• One local store at ~/.monet — plain SQLite you can read, back up, and export.
• Embeddings run on-device. No accounts, no API keys, no cloud.
• The memory engine runs locally; after the first model download, Monet works offline.
• monet dashboard is read-only and offline — run it with wifi off and your memory graph still renders.

Setup may download the local embedding model once. Nothing about your code is uploaded by Monet.

monet dashboard

Preview what Monet remembers.

Run monet dashboard for a local, force-directed view of your memory — every concept, sized by importance, and the links between them. Read-only, fully offline.

Demo dashboard. Your own memory graph runs locally with monet dashboard.

What Monet is

Working memory for your projects — not a universal memory API.

Monet isn't trying to remember everything for every app. It does one thing well: keep your coding agent's project state sharp across sessions, locally. That focus is the point.

Give your agent a memory.

Star with-monetmonet dashboard

Free to use for any project. with-monet, the harness, is open source.