Quickstart
This quickstart wires the MemMesh memory engine into your editor as an MCP server. The recommended path is fully local: one command installs the engine and points your editor at it over a local stdio connection. No API key, no server, no Docker.
No API key needed for local use. The local engine stores everything in a
SQLite file on your machine (~/.thinkfleet-memory/memory.db) and runs on the
Free tier out of the box.
Prerequisites
- An AI editor that speaks MCP — Claude Code, Cursor, Windsurf, or Codex.
- Node.js 18+ (to run the
npxinstaller).
1. Install with one command
The @thinkfleet/memmesh npm package is a thin shim: it resolves the
thinkfleet-memory engine binary and runs the installer, which writes the MCP
server config into your editor.
npx @thinkfleet/memmesh installRun bare, it wires every detected tool. Target one with --tool:
npx @thinkfleet/memmesh install --tool claude-codeSwap claude-code for cursor, windsurf, or codex. The installer writes the
MCP server entry and the agent teaching skill, preserving any servers you already
have.
The prebuilt binary release channel isn’t live yet. Until it is, the shim
can’t auto-download the engine — point it at a thinkfleet-memory binary you
already have, or build one from source (see below):
THINKFLEET_BINARY=/path/to/thinkfleet-memory npx @thinkfleet/memmesh installIf you have the binary on your $PATH, you can also run it directly:
thinkfleet-memory install --tool claude-code.
Run the installer with --dry-run to preview the exact config that would be
written without touching any files. See the
Installer & doctor page for all flags and the
doctor diagnostic.
2. Build the engine from source (until releases ship)
While the release channel is being wired up, build the engine from the
thinkfleet-memory-engine source tree with a Rust toolchain (cargo):
cargo build --release
# binary at ./target/release/thinkfleet-memoryPut that path in THINKFLEET_BINARY (or on your $PATH) before running the
installer.
3. Restart and verify
Restart your editor so it picks up the new MCP server. The MemMesh memory tools
(memory_observe, memory_save, memory_search, memory_recall, and more)
should now appear in the tool list — see the full
MCP tool reference.
Check the install any time with the built-in diagnostic:
thinkfleet-memory doctor4. Use it
In your editor’s chat, try:
- “Remember that the EU region requires data residency in eu-west-1.” — writes a memory.
- “What rules do we have for refunds?” — searches your project’s memories.
On Claude Code, an auto-observe hook also runs thinkfleet-memory observe on
every prompt, so the engine captures facts without you asking. See the
Claude Code integration for details.
Hosted access
Mesh Router (hosted) — coming soon. A managed, multi-tenant memory endpoint with per-project auth is in development. Until it ships, run the local engine (or self-host the stack on your own infrastructure).