DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to run a Docker MCP server for Claude Code memory with MemPalace

Pull the multi-arch MemPalace image and wire it into an MCP client as a stdio server, with the palace persisted in a Docker volume.

Short answer: Run docker pull ghcr.io/mempalace/mempalace:latest (multi-arch, runs natively on Apple Silicon); then start the MCP server over stdio with docker run -i --rm -v mempalace-data:/data ghcr.io/mempalace/mempalace (keep the -i flag, JSON-RPC needs stdin); add a MemPalace entry under mcpServers in your MCP client config with command docker and the same run arguments, mounting your .claude/projects folder read-only at /transcripts. All steps.

Remote script. This pulls and runs a container image from ghcr.io; the README warns about impostor sites, so pull only the official ghcr.io/mempalace image.

Who it is forA developer who wants agent memory over MCP without installing a local Python toolchain.

The problemInstalling chromadb, numpy and grpcio into a global Python can conflict with other tools, and some Pythons block it with PEP 668 errors.

The outcomeAn MCP server your agent can call for palace reads and writes, with the palace, config and embedding model kept under /data in a named volume.

What you need first

  • Python 3.10 or newer
  • uv or pipx recommended for the CLI (plain pip only inside a virtualenv)
  • A vector-store backend (ChromaDB by default, no configuration)
  • About 300 MB of disk for the embedding model
  • Docker only if you use the container image

Steps

  1. Run docker pull ghcr.io/mempalace/mempalace:latest (multi-arch, runs natively on Apple Silicon)
  2. Start the MCP server over stdio with docker run -i --rm -v mempalace-data:/data ghcr.io/mempalace/mempalace (keep the -i flag, JSON-RPC needs stdin)
  3. Add a MemPalace entry under mcpServers in your MCP client config with command docker and the same run arguments, mounting your .claude/projects folder read-only at /transcripts
  4. Use a real absolute path for the transcripts mount, because ~ and $HOME are not expanded by every MCP client
  5. Test retrieval from the CLI with docker run --rm -v mempalace-data:/data ghcr.io/mempalace/mempalace search "why GraphQL"

Prompt for your agent

Set up MemPalace/mempalace as a stdio MCP server for Claude Code using the Docker image: run `docker pull ghcr.io/mempalace/mempalace:latest`, then add a mempalace entry to my MCP config that runs the image with -i, the mempalace-data volume on /data and my Claude Code projects folder mounted read-only at /transcripts, using an absolute path. Show me the config before saving it.

Caveat. The first command that needs embeddings downloads the model into /data (about 80 MB for minilm, about 300 MB for embeddinggemma), so the first call is slow and needs network. On Linux the image runs as uid 1000, so a 0700 mounted folder fails with a PermissionError.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for MemPalace/mempalace and "Run a Docker MCP server for Claude Code memory"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer