Release archives
Every tag vX.Y.Z publishes five archives and a checksum file on the
releases page:
| Platform | Archive |
|---|---|
| macOS, Apple silicon | memors-mcp_X.Y.Z_darwin_arm64.tar.gz |
| macOS, Intel | memors-mcp_X.Y.Z_darwin_amd64.tar.gz |
| Linux, x86-64 | memors-mcp_X.Y.Z_linux_amd64.tar.gz |
| Linux, ARM64 | memors-mcp_X.Y.Z_linux_arm64.tar.gz |
| Windows, x86-64 | memors-mcp_X.Y.Z_windows_amd64.zip |
| All | checksums.txt (SHA-256) |
Each archive contains the binary, LICENSE, README.md and SKILL.md.
The optional macOS menu-bar app comes as one more asset, memors-tray_X.Y.Z_darwin_universal.zip,
with its own .sha256 file; see MemoRS Tray.
Install
VERSION=1.4.0
ARCH=darwin_arm64 # see the table above
curl -LO https://github.com/kKEo/memors/releases/download/v${VERSION}/memors-mcp_${VERSION}_${ARCH}.tar.gz
curl -LO https://github.com/kKEo/memors/releases/download/v${VERSION}/checksums.txt
shasum -a 256 -c --ignore-missing checksums.txt # Linux: sha256sum -c --ignore-missing
tar xzf memors-mcp_${VERSION}_${ARCH}.tar.gz
install -m 0755 memors-mcp ~/.local/bin/memors-mcp # any directory; note the absolute path
memors-mcp version
memors-mcp version prints the build version, the MCP protocol version (2026-07-28), the Go
version and the model directory.
macOS Gatekeeper. A binary downloaded by a browser carries a quarantine attribute and is blocked on first run. Clear it with
xattr -d com.apple.quarantine ~/.local/bin/memors-mcp, or download withcurl, which does not set it.
MCP registry
The server is listed in the MCP registry as io.github.kKEo/memors. Its server.json
points at the release archives with their checksums, so registry-aware clients can install it
directly.
What the binary needs at runtime
- No shared libraries, no CGo, no interpreter. (This is the memors-mcp binary. MemoRS Tray, the separate macOS app, links AppKit and WebKit.)
- A writable
MEMORS_HOME, by default~/.memors-mcp, and a writable~/.cache/memors-mcp/models. - About 200 MB of disk for the default model, plus the knowledge-base files. See Capacity and performance.
- Network access to
huggingface.coonce per model, unless you pre-seed the cache.