Agent Memory MCP — Claude Code MCP server: what it does
Keeps engineering context, project docs, and incident logs in a local SQLite database. Useful for platform and DevOps teams tracking operational decisions across coding sessions.
Open on GitHubWhat Agent Memory MCP does
Instead of storing flat notes, the server organizes context across four distinct categories: episodic, semantic, procedural, and working memory. Records and vector indexes reside inside a local SQLite database. The engine reads directly from disk. This avoids memory spikes when handling large knowledge bases.
Retrieval combines vector embeddings with BM25 keyword matching while prioritizing recent and verified records. Built-in RAG indexes runbooks, RFCs, changelogs, and infrastructure files. It scrubs secrets before ingestion. For Claude Code, lifecycle hooks automatically capture takeaways at session completion and save checkpoints before context compression.
Switching embedding models requires running agent-memory-mcp reembed to prevent mismatched vectors. The server communicates through stdio or HTTP/JSON-RPC. Offline setups rely on Ollama.
What it helps with
- preserve architectural decisions and context between agent sessions
- surface relevant runbooks and postmortems during incident triage
- index local technical documentation with automatic secret redaction
- maintain local project memory in SQLite without external databases
Example prompts for the agent
- Search runbooks and past postmortems to see why HPA was disabled on this service.
- Save the decision to update Terraform parameters into procedural memory.
- Check the project memory for recent infrastructure changes that could cause regressions.
How to install Agent Memory MCP
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Agent Memory MCP
How do you install Agent Memory MCP for Claude Code?
The tool is distributed as prebuilt binaries, a Homebrew formula, a Docker image, or from source. To connect it to Claude Code, register the executable in your MCP client configuration using stdio mode.
What does Agent Memory MCP do?
It provides a local SQLite-backed context layer combining typed memory, document search, and repository tools. AI coding agents use it to reference architectural decisions, runbooks, and changelogs across sessions.
Can Agent Memory MCP run completely offline?
Yes, you can enable local-only mode. The server directs embedding generation to a local Ollama instance, avoiding calls to hosted AI providers.
