state-trace

A local MCP server giving coding agents a bounded working-memory graph with typed nodes and current-versus-stale state, backed by a SQLite or JSON file.

What you can do

Store and recall working-memory entries as typed nodes in a persistent graph, so an assistant can keep context across a coding session rather than losing it between turns. Track whether a remembered item is current or stale, helping the assistant avoid acting on outdated state. Scope reads and writes to a namespace, such as a repository slug, while nodes without a namespace stay visible to every namespace. Keep memory within a configurable capacity budget, and choose a SQLite or JSON file as the backing store.

When to use it

Worth adding when you want a coding agent to carry structured, persistent working memory across a session or across repositories, and to distinguish fresh state from stale state instead of re-deriving context each time.

Requirements

Runs as a local stdio process installed from the PyPI package state-trace, so a Python environment is needed. Optional environment variables: STATE_TRACE_STORAGE_PATH for the backing SQLite or JSON file (defaults to a file under the user's home directory), STATE_TRACE_NAMESPACE to scope reads and writes, and STATE_TRACE_CAPACITY_LIMIT for the memory budget (default 256). No authentication is declared.