How to use the Claude-Mem Memory Search agent skill (mem-search)

What the mem-search skill makes the agent do

Claude-Mem stores short timestamped observations collected from your previous sessions across agents such as Claude Code, Cowork and Codex. When you ask about earlier work, the skill tells the agent to search that memory on cmem.ai and answer in plain language from the matching observations, rather than guessing or asking you to repeat yourself.

Install the skill from the ZIP or the Claude apps

Download the ZIP for this skill from this page and unzip it. For a personal setup, place the unzipped skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, open Settings, then Capabilities, then Skills, and upload the skill there.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills

Meet the requirements before you start

The search cannot run without these three things. The skill runs a bundled script that needs Node.js. You also need a cmem.ai account with a Claude-Mem API key configured in the plugin, and the claude-mem plugin installed, which provides the CLAUDE_PLUGIN_ROOT location and its scripts folder.

• Node.js (runs the bundled cmem-hook.mjs script)
• A cmem.ai account with a Claude-Mem API key set in the plugin
• The claude-mem plugin installed, providing CLAUDE_PLUGIN_ROOT and its scripts folder

Trigger the skill with natural requests

You do not need a special command. The skill is meant to be used when you ask things such as searching memory, asking what the agent remembers about a topic, checking claude-mem, finding past observations, or asking what was done last session. Name the project, file, decision or task so the agent has a clear query.

• "Search memory for the auth refactor decision."
• "What did we do last session on the billing export?"
• "What do you remember about the retry logic in api/client.ts?"
• "Check claude-mem for the error we hit with the migration."
• "When did we switch the build tool? Use my memory."

How the agent searches and answers

The skill runs the bundled CLI with your query. It follows a progressive method: a broad index pass of one to three keyword searches (project names, file names, error strings), then a narrow pass using the most specific terms found, with a smaller limit. The agent then synthesizes an answer and quotes timestamps when you ask when something happened. It is instructed not to dump raw output.

• Index pass: 1–3 broad keyword searches, skimming titles and summaries
• Narrow pass: re-search with exact terms and a smaller limit
• Answer: a plain-language summary, with timestamps for 'when' questions

Check the status when searches fail

If a search returns nothing or errors, the skill can run a status check. A missing API key means the plugin still needs your cmem.ai key added to its configuration. A 404 on /api/hooks/context means the newer Pro endpoint is not deployed, but search still works through /api/mcp.

Safety notes: scripts and network use

This skill runs a script and uses the network. Every search query is sent to cmem.ai, so do not put secrets such as passwords or keys into queries. Some filters, such as date ranges or type, may be silently ignored by the cloud API, so narrow results with keywords rather than filter flags. If your project is sensitive or confidential and you do not want queries going to a third-party cloud service, do not use this skill.

• Queries are sent to cmem.ai
• Never write secrets into search queries
• Date and type filters may be silently ignored; prefer keyword narrowing