DEDUPEDWho it is forA developer who switches between AI coding tools and keeps re-explaining the same project context.
The problemEach tool starts without the context gathered in the others, so work gets repeated.
The outcomeEvery connected MCP client searches the same index and stores durable memories in one instance directory.
What you need first
- Python 3.11 or newer
- git and a terminal
- An embedding endpoint that supports /embeddings, plus a chat endpoint (may be a local one)
- A notes or project docs directory to index
Steps
- Complete the setup in the quick start so the instance directory contains config.toml.
- Add the Loci server entry to each MCP client config, with command set to loci-mcp and env LOCI_HOME set to the absolute path of the instance directory.
- If a client cannot find loci-mcp, replace the command with the absolute path to the executable.
- Restart the client and call brain_remember to store a memory, then brain_search to retrieve it from another client.
Prompt for your agent
Using the IvenKooLab/loci repo, add a loci entry to the MCP config of <client-name> with LOCI_HOME set to <absolute-instance-path>. Expected output: the client lists brain_search and brain_remember tools. Verify that a memory stored by one client is returned by brain_search in a second client.
Caveat. The README says the instance directory must contain config.toml, and every client must point at the same LOCI_HOME for the sharing to work.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for IvenKooLab/loci and "Share one memory store across MCP-connected AI tools"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer