memtomem-stm
MCP proxy that compresses and caches tool responses and surfaces prior-session memory.
Tested on Linux
Use it
Linux x86_64
You need first
- Runtimes:
python 3.12 and 3.13 supported
Install via uv tool (candidate 1)
uv tool install memtomem-stm
Where the README and reality differ
- README: call demo__demo_search with {"query": "cache"}
What worked: call with '{"topic": "cache"}' as advertised by tools/list (arg name is topic)
Check that it works
mms versionThe output should contain “memtomem-stm 0.7.2”.
First thing to try
mms versionAdd it to your agent
claude mcp add memtomem-stm -- mmsJSON for Claude Desktop, Cursor and other clients
{
"mcpServers": {
"memtomem-stm": {
"command": "mms",
"args": []
}
}
}Run mms version to confirm the install, then point your MCP client at memtomem-stm so it sits in front of your MCP servers.
What it does
- Compress tool responsesNot shown working
- Cache repeated tool callsNot shown working
- Bring up context from earlier sessionsNot tested
How far pickfit got
Linux x86_64
- Installs
- Connects
- 0 of 3 features work
- Works offline
Good for
- compress and cache repeated MCP tool responses
- surface prior-session memory from memtomem LTM
- add compression and observability in front of any MCP server
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
An MCP client called one of its tools
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 0.7.2
An MCP client handshake over stdio listed 6 tools, and a call to demo__demo_search returned a result.
Differences the test machine hit
- README: mms init --demo --client auto as no-network first-success path
In the test: run non-interactively with piped stdin and --no-validate, e.g. printf 'n\n' | mms init --demo --client auto --no-validate - README: three entry points resolve to same CLI and MCP server entrypoint
In the test: bare `mms` with piped stdin boots the proxy as MCP server, so probe launch is simply `mms`