Open source · Apache 2.0

Shared memory for
your AI agents

One knowledge base for every agent on your team. Detects contradictions automatically. Four MCP tools. Zero config.

$ pip install engram-mcp && engram serve --http
or, no install needed:
$ uvx engram-mcp@latest serve --http

Python 3.11+ · Runs on localhost:7474 · No API keys

Up and running in 3 steps

Like Claude Code — install, run, connect. That's it.

1

Install

One pip install. No Docker, no database setup, no API keys.

pip install engram-mcp
2

Run

Start the server. Dashboard at localhost:7474.

engram serve --http
3

Connect

Add to your MCP client config. Pick your transport:

{
  "mcpServers": {
    "engram": {
      "url": "http://localhost:7474/mcp"
    }
  }
}
{
  "mcpServers": {
    "engram": {
      "command": "uvx",
      "args": ["engram-mcp@latest"]
    }
  }
}

Four tools. That's the entire API.

Your agents call them automatically through MCP.

engram_query

Pull what your team's agents collectively know. Structured facts, ranked by relevance.

engram_commit

Persist a verified discovery. Append-only, timestamped, immediately shared.

engram_conflicts

Surface contradictions between facts. Reviewable, resolvable, auditable.

engram_resolve

Settle disagreements. Pick a winner, merge, or dismiss false positives.

Works everywhere

Any MCP-compatible client. No vendor lock-in.

Claude Code
Cursor
Windsurf
Kiro
VS Code
Any MCP Client