agentmem — Claude Code MCP server: what it does, how to install
Keeps an agent's notes across sessions and tracks which ones still hold. Useful when your assistant keeps repeating old mistakes.
claude mcp add agentmem -- uvx quilmemWhat agentmem does
agentmem is an MCP server that gives coding agents a shared memory. Every entry carries a status: hypothesis, active, validated, deprecated or superseded. Recall ranks validated and active entries above the rest and leaves deprecated and superseded ones out entirely.
The server also looks for entries that disagree. It measures topic overlap, matches negation at sentence level, and keeps duplicates apart from real contradictions. A separate check flags stale entries by age, by a missing source file, or by a hash that no longer matches. One more function scores the whole memory store from 0 to 100 based on conflicts, the share of stale entries and whether anything has been validated.
It runs through uvx quilmem, so uvx has to be on your machine. The Python API comes from the quilmem package, the MCP mode from quilmem[mcp]. Memory lives in a local database file, no external service involved.
What it helps with
- stop the agent from repeating mistakes made in earlier sessions
- tell a current rule apart from an outdated one
- surface decisions in memory that contradict each other
- flag entries whose source file changed or disappeared
- fit the best entries into a token budget when building context
Example prompts for the agent
- Save to memory: never apply loudnorm to the final mix, status validated
- Find conflicts in this project's memory and show which entries contradict each other
- Build context for the narration track task, stay within 2000 tokens
How to install agentmem
Platform CLI
claude mcp add agentmem -- uvx quilmemuvx runs the command named after the package; if the README names another one, use uvx --from quilmem <command>
Platform CLI
codex mcp add agentmem -- uvx quilmemuvx runs the command named after the package; if the README names another one, use uvx --from quilmem <command>
Config file
{
"mcpServers": {
"agentmem": {
"command": "uvx",
"args": [
"quilmem"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from quilmem <command>
Config file
{
"servers": {
"agentmem": {
"type": "stdio",
"command": "uvx",
"args": [
"quilmem"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from quilmem <command>
Platform CLI
gemini mcp add agentmem uvx quilmemuvx runs the command named after the package; if the README names another one, use uvx --from quilmem <command>
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 agentmem
How do I install agentmem in Claude Code?
Run claude mcp add agentmem -- uvx quilmem, then restart your editor.
What does agentmem do?
It stores an agent's notes with a type and a status, searches them by text, and assembles context within a token limit. It also detects conflicting and stale entries and scores the health of the memory store.
How is agentmem different from a plain memory store?
Entries move through a lifecycle from hypothesis to validated or superseded, so recall skips what is no longer true. Conflict detection, staleness checks and health scoring are built in.
