MemPalace MCP
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
- 59,837 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads project files and session transcripts
- Hooks write data to the local database automatically
Install
- Works in
Manual install
npx skills add MemPalace/mempalaceInstalls only the skills; the agent then installs the CLI and MCP server via the mempalace skill.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads project files and session transcripts
- Hooks write data to the local database automatically
This is third-party code. Review the repository files before installing.
What it does
MemPalace puts your agent conversation history and project files into a local index without summarizing or extracting facts. Data is structured: people and projects are wings, topics are rooms and original text sits in drawers, so searches can be scoped. The MCP server gives the agent tools for reading and writing memory, a temporal entity graph, agent diaries and task handoffs. Hooks for Claude Code, Codex and Cursor save sessions during work and before context compaction. The repository also ships skills for guided setup, search-before-answer recall and task delegation.
Who it is for. For people who work with agents on projects over time and want past decisions and discussions retrievable without cloud services.
Good fit when
- The agent forgets why a decision was made in the project
- You need to keep Claude Code sessions that otherwise expire after 30 days
- Several machines or agents should share memory through a hub
Not a fit when
- You want a short fact summary rather than verbatim storage
- Disk space is tight or Python 3.9+ is unavailable
Example request
Search memory for why we switched to GraphQL and take it into account in the planMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
MEMPALACE_BACKEND- Vector backend: chroma, sqlite_exact, milvus, qdrant, pgvector and others.
Limitations
Requires Python 3.9+ and about 300 MB for the embedding model; the first model download needs network access. Native Termux install is not supported. The only official sources are GitHub, PyPI and mempalaceofficial.com; the authors warn about impostor sites.
How to disable. Remove the mempalace MCP server from the agent config or the plugin via /plugin, then run uv tool uninstall mempalace. Palace data is stored locally and removed separately.
FAQ
Do I need an API key?
No, core search runs locally without an LLM or keys. Embeddings can be moved to your own OpenAI-compatible server.
How do I import old Claude Code sessions?
Run mempalace mine ~/.claude/projects/ --mode convos.
Discussion
No questions or reviews yet.
Related
MCP serversA local memory MCP server for coding agents: records issues, attempts, fixes and decisions and warns before repeating a failed approach
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
An MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Self-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
README in short
The README explains the MemPalace storage model and offers three install paths: skills with agent-guided setup, uv or pipx, and a Docker image with an MCP config example. It covers pluggable backends (ChromaDB by default, SQLite, Milvus, Qdrant, pgvector), the mine, search and wake-up commands, benchmark results with reproduction steps, the knowledge graph and auto-save hooks. MIT licensed.