Copy & run

Commands & setup

Everything you need to clone it, run it, wire it into an IDE, and verify the claims yourself. Filter by CLI (things you run) or Reference (things to know).

From the guide

Terminal · Claude CodeRun the MCP server (Python)

python3 -m prime_memory_folding.mcp_server

Terminal · Claude CodeRun it through the Node launcher

node bin/prime-memory-folding-mcp.js

Quick start

Terminal · Claude CodeClone & run the tests

git clone https://github.com/coryhubbell/prime-memory-folding.git
cd prime-memory-folding
python3 -m unittest discover -s tests

Terminal · Claude CodeEncode a domain record

python3 -m prime_memory_folding encode architecture --tags '["technical","standalone"]'

Terminal · Claude CodeRemember something with a vector

python3 -m prime_memory_folding remember "Prime filters are for hot deterministic recall." --tags '["prime","technical"]' --vector '[1,0,0]' --importance 0.9

Terminal · Claude CodeQuery by tag

python3 -m prime_memory_folding query --tags '["technical"]'

Terminal · Claude CodeRun a folding pass

python3 -m prime_memory_folding fold

MCP & IDE

Terminal · Claude CodeStart the MCP server

python3 -m prime_memory_folding.mcp_server

Terminal · Claude CodeOr the Node launcher

node bin/prime-memory-folding-mcp.js

Claude DesktopThe five MCP tools

prime_memory_encode
prime_memory_store
prime_memory_query
prime_memory_fold
prime_memory_stats

Verify it yourself

Terminal · Claude CodeRun the filter benchmarks

python3 benchmarks/benchmark_filters.py

Claude DesktopThe four benchmark paths

domain_filter_unsorted  -> clear constant-factor win
tag_predicate_only      -> faster in the bundled benchmark
tag_filter_unsorted     -> parity to modestly faster
tag_query_sorted        -> parity to modestly faster