Compare

total-agent-memory vs Claude Code's built-in memory

Claude Code already has CLAUDE.md files and auto memory. What they do, where they stop (per repository, no search by meaning, one tool only), and what an MCP memory server adds on top. They work together, not instead of each other.

Facts about Claude Code built-in memory were checked on October 9, 2026 against its public source. Facts about total-agent-memory are current for the version in the site header.

What Claude Code already gives you

Claude Code has two built-in mechanisms, both loaded at the start of every session:

CLAUDE.md files. Instructions you write: build commands, conventions, rules. They live in the repository, in your home directory, or in a managed policy file, and the whole file is read every session.
Auto memory. Notes Claude writes for itself about your preferences, your corrections and project context it cannot derive from the code. Each repository gets a directory under ~/.claude/projects/<project>/memory/ with a MEMORY.md index and one file per topic. The first 200 lines or 25 KB of the index load at session start; topic files are read on demand.

This is good and you should use it. For a single repository and a single tool it covers the basics.

Where it stops

Per repository. Auto memory is keyed by the git repository. A lesson learned in one project is not there in the next one.
Per tool. Cursor, Codex CLI and other agents do not read it. If you switch tools, the memory stays behind.
Per machine. Files are local and not shared across machines, and there is no team layer.
No search by meaning. The index is read top to bottom. Nothing finds “the time we chose pgvector over Chroma” from the question “which vector store did we pick”.
No structure. A decision, a fix, a fact and a session summary are all lines in a markdown file; there is no type, no project tag, no validity date for a fact that changed.

What total-agent-memory adds

One memory for all your projects and all your agents: Claude Code, Codex CLI, Cursor, Claude Desktop, Cline, Continue, Windsurf, Gemini CLI, OpenCode and Aider share one database.
Search by meaning as well as by words, fused with a knowledge graph, with no API key and no model call on the search path.
Typed records: decisions with their reasons, solutions, facts with validity dates, lessons, session summaries with next steps.
Session start and end tools, so an agent picks up where the last one stopped.
A team server with personal, team and shared scopes for companies.

Use both

Keep CLAUDE.md for rules and instructions: it is the right place for “run the tests before you commit” and “handlers live in src/api/handlers/”. Let TAM hold knowledge: why a decision was made, how a bug was fixed, what the staging port is, what to avoid next time. The Claude Code plugin installs the MCP server, a memory-protocol skill that tells Claude when to save and recall, and capture hooks, in one step:

/plugin marketplace add vbcherepanov/total-agent-memory
/plugin install total-agent-memory@vbcherepanov

Then restart Claude Code and check /mcp shows memory as connected. The full walkthrough is in Persistent memory for Claude Code.

Facts about Claude Code’s memory come from its documentation on the date above.