How do I see what my AI agent remembers?

In Claude Code, run /memory. It lists your CLAUDE.md files and lets you open the auto memory folder, where Claude's notes are plain markdown you can read, edit or delete. For an MCP memory server, open its file or call its read tool. Other agents keep memory in their own places.

Claude Code: two kinds of memory

Claude Code carries knowledge across sessions in two ways. CLAUDE.md files hold instructions that you write. Auto memory holds notes Claude writes itself from your corrections and preferences. Both are loaded at the start of each session, and auto memory loads its first 200 lines or 25KB.

Auto memory lives in ~/.claude/projects/<project>/memory/. Every worktree and subfolder of one repository shares one memory folder. You can move it with the autoMemoryDirectory setting.

See, edit and delete it

  1. Run /memory in a session. It lists your CLAUDE.md, CLAUDE.local.md and other memory files, and offers to open the auto memory folder.
  2. Select a file to open it in your editor. Everything is plain markdown, so you can fix a wrong note or delete a file.
  3. Run /context to check which CLAUDE.md and rules files actually loaded in this session.
  4. Toggle auto memory from /memory, which saves autoMemoryEnabled to ~/.claude/settings.json.
  5. Ask Claude to remember something, or to add it to CLAUDE.md instead. The words you use decide which one it writes to.

MCP memory servers

If you use the reference knowledge graph server, the memory is a JSONL file at the path in MEMORY_FILE_PATH, which you can open in any editor. Its tools include read_graph for the whole graph, search_nodes and open_nodes for parts of it, and delete tools for entities, observations and relations. Ask the agent to run them and it will show you what it holds.

Other agents

Each tool stores memory differently. ChatGPT, Cursor, Codex and others each have their own settings and files, and the names change between versions. Look in the tool's settings for a memory or rules page, and in the project for instruction files such as AGENTS.md. If the tool offers no way to read its memory, treat what it remembers as unknown and keep your important facts in a file you control.

What to look for when you read it

Read memory the way you would review a colleague's notes. Look for facts that were true once and are not now, such as a renamed folder or a changed command. Look for preferences stated as rules, such as always use tabs, that you only meant for one task. Look for anything private that should not be there, and for two notes that disagree, because the agent may pick either one.

When you fix an entry, fix it at the source. If a CLAUDE.md or AGENTS.md line says the wrong thing, the agent will write it back into memory next session. Edit the file the fact came from, then the note that copied it.

Audit habits that help

Review memory now and then, and delete what is wrong or stale, since an agent treats a remembered fact as true. Keep secrets out of it. Prefer small, dated, specific facts to long summaries. Memory also shapes behavior, so when an agent does something odd, check memory before blaming the prompt.

Where a NodePress node fits

Each tool's memory is a separate pile. In a NodePress node, agents share one memory, a graph, plus a thread where every step is written in one readable place, so you can see what each agent did and what it added. Agents join with one pasted line through the remote MCP server at https://nodepress.ai/mcp. The node runs on your own computer.

Do it with a NodePress node

  1. Name a node on nodepress.ai. It is free to start, with no sign-up for the first node.
  2. Paste the one line into each agent that should share memory.
  3. Open the node's thread to read each step.
  4. Look through the shared memory for facts that are wrong or old.
  5. Correct or remove them so every agent stops using them.

If you use one agent and its own /memory view shows what you need, you do not need a node to inspect it.

Questions people ask

How do I see what my AI agent remembers?

In Claude Code, run /memory. For an MCP memory server, open its file or call read_graph. Other tools have their own memory or rules pages.

Where does Claude Code store memory?

Auto memory is in ~/.claude/projects/<project>/memory/, shared by all worktrees of the repository. CLAUDE.md files live in your project and home folder.

How do I delete my agent's memories?

In Claude Code, open /memory and delete or edit the files, which are plain markdown. For an MCP graph, use the delete tools or edit the JSONL file.

Can I turn memory off?

In Claude Code, /memory has a toggle for auto memory, saved as autoMemoryEnabled in your user settings.

How can I see what my agents did and why?

Read each agent's own session log, or use a shared thread where every step is recorded in one place, as in a NodePress node.

Related

Sources

NodePress gives your AI agents one shared place on your own computer. Free to start, no sign-up. Make your first node.