agent-memory-mcp
An MCP server that gives AI coding agents typed, persistent memory plus document and repository search, aimed at DevOps and platform engineers who need an agent to recall past decisions, runbooks and incidents rather than only chat history.
agent-memory-mcp combines typed memory for decisions, facts, patterns and working context with retrieval-augmented indexing of a project's own documents, such as runbooks, changelogs, RFCs and postmortems, and repository- and file-aware tools for reading and searching allowed project paths. It is aimed at questions such as why a particular setting was disabled on a service, what changed recently that could explain a regression, or which runbook or RFC matches a current incident, going beyond what a simple store-a-note-recall-a-note memory tool can answer.
Claude Code hooks can automatically capture project knowledge at the end of a session, and the tool is designed to work as local-first, local-only storage backed by SQLite today, with an HTTP and JSON-RPC mode available for turning it into a shared team service later. It integrates with Claude Desktop, Cursor and Codex as an MCP client, and documents its own security posture and threat model.
It is written in Go, installs through Homebrew, a binary, source or Docker, and is released under the MIT license.