Pseudolife Memory — Claude Code MCP server: long-term agent memory
Gives a coding agent memory on disk: notes, canonical facts and a knowledge graph that outlive context compaction and new tasks.
claude mcp add pseudolife-mcp -- uvx pseudolife-mcpWhat Pseudolife Memory does
The server keeps an associative store where notes are ranked by hybrid retrieval. Conflicts get flagged instead of silently overwritten, and whole-note replacement is a deliberate step. On top sits a fact layer: one current value per entity.attribute slot, with corrections superseding the old one while the version history stays around.
A bundled local extractor or any OpenAI-compatible endpoint consolidates the memory stream into facts and a graph in the background. Successes, dead ends and your corrections turn into do/avoid hints shown at session start. The Cortex Console web UI lets you watch this happen, with cited facts, session episodes and document RAG.
Setup is pip install "pseudolife-mcp[lite]" plus registering the server in your client — no Docker, no database. Codex needs more than the add command: set startup_timeout_sec = 240, tool_timeout_sec = 240 and required = true in ~/.codex/config.toml, because a cold daemon can wait up to 180 seconds while Codex allows 10 by default. Approve memory_message there too, or a woken task stalls on a prompt. Treat those numbers as starting budgets.
What it helps with
- keep decisions alive between agent sessions
- retain a fact after context compaction
- separate a corrected value from the outdated one
- build a knowledge graph out of accumulated notes
- surface hints from past wins and dead ends
Example prompts for the agent
- Remember that this project builds with
uvxand bring it up next session. - What did we already decide about the auth scheme?
- Show the facts for
billing.serviceand their edit history.
How to install Pseudolife Memory
Platform CLI
claude mcp add pseudolife-mcp -- uvx pseudolife-mcpuvx runs the command named after the package; if the README names another one, use uvx --from pseudolife-mcp <command>
Platform CLI
codex mcp add pseudolife-memory -- uvx pseudolife-mcpuvx runs the command named after the package; if the README names another one, use uvx --from pseudolife-mcp <command>
Config file
{
"mcpServers": {
"pseudolife-memory": {
"command": "uvx",
"args": [
"pseudolife-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from pseudolife-mcp <command>
Config file
{
"servers": {
"pseudolife-memory": {
"type": "stdio",
"command": "uvx",
"args": [
"pseudolife-mcp"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from pseudolife-mcp <command>
Platform CLI
gemini mcp add pseudolife-memory uvx pseudolife-mcpuvx runs the command named after the package; if the README names another one, use uvx --from pseudolife-mcp <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 Pseudolife Memory
How do I install Pseudolife Memory in Claude Code?
Run pip install "pseudolife-mcp[lite]", then claude mcp add pseudolife-memory -- uvx pseudolife-mcp. No Docker or database is involved.
What does Pseudolife Memory do?
It is an MCP server for long-term memory: notes, canonical facts and a knowledge graph stored on disk so they survive context compaction and fresh tasks.
What has to be configured for Codex?
In the [mcp_servers.pseudolife-memory] table of ~/.codex/config.toml, add startup_timeout_sec = 240, tool_timeout_sec = 240 and required = true, and approve the memory_message tool.
