engram · kael-bit — Claude Code MCP server: what it does, how to install
Hierarchical memory engine with three-stage retention and decay for AI agents. Keeps essential project context alive across session restarts.
claude mcp add --env ENGRAM_URL=<ENGRAM_URL> --env ENGRAM_API_KEY=<ENGRAM_API_KEY> --transport stdio kael-bit-engram -- npx -y engram-rs-mcpWhat engram does
The server manages agent memory across buffer, working, and core tiers. Instead of relying on hardcoded heuristics, an LLM evaluates whether a memory merits promotion into long-term storage. Retention rates depend directly on content type. Episodic records fade fastest, whereas procedural workflows persist over extended epochs. Recurring access triggers a score reinforcement.
Under the hood, the engine compiles into a single Rust binary using SQLite, bypassing separate vector databases. It clusters related entries into subject branches and merges duplicated facts without manual tagging.
The MCP connection will fail without a running backend supplying ENGRAM_URL and ENGRAM_API_KEY. The interactive setup script also halts until you provide embedding credentials.
What it helps with
- retain critical decisions across separate coding sessions
- deprecate obsolete context automatically through decay cycles
- cluster stored memories into searchable topic hierarchies
- fetch deployment guidelines using specific trigger tags
Example prompts for the agent
- Store a rule that unit tests must pass before any production deployment.
- Retrieve all stored lessons tagged for database migrations.
- Restore working context from previous sessions using engram.
How to install engram
Platform CLI
claude mcp add --env ENGRAM_URL=<ENGRAM_URL> --env ENGRAM_API_KEY=<ENGRAM_API_KEY> --transport stdio kael-bit-engram -- npx -y engram-rs-mcpReplace <…> with your own values
Platform CLI
codex mcp add kael-bit-engram --env ENGRAM_URL=<ENGRAM_URL> --env ENGRAM_API_KEY=<ENGRAM_API_KEY> -- npx -y engram-rs-mcpReplace <…> with your own values
Config file
{
"mcpServers": {
"kael-bit-engram": {
"command": "npx",
"args": [
"-y",
"engram-rs-mcp"
],
"env": {
"ENGRAM_URL": "<ENGRAM_URL>",
"ENGRAM_API_KEY": "<ENGRAM_API_KEY>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"kael-bit-engram": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"engram-rs-mcp"
],
"env": {
"ENGRAM_URL": "<ENGRAM_URL>",
"ENGRAM_API_KEY": "<ENGRAM_API_KEY>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e ENGRAM_URL=<ENGRAM_URL> -e ENGRAM_API_KEY=<ENGRAM_API_KEY> kael-bit-engram npx -y engram-rs-mcpReplace <…> with your own values
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 engram
How to install engram in Claude Code?
Run claude mcp add --env ENGRAM_URL=<ENGRAM_URL> --env ENGRAM_API_KEY=<ENGRAM_API_KEY> --transport stdio kael-bit-engram -- npx -y engram-rs-mcp.
What does the engram MCP server do?
It connects coding agents to a tiered memory architecture comprising buffer, working, and core stages. Knowledge clusters automatically by topic, merges duplicates, and fades over time when unreferenced.
Does engram delete unaccessed memories permanently?
Memories drop to a minimum retention floor of 0.01 rather than vanishing completely. They remain retrievable through exact queries, and procedural instructions decay slower than temporary episodic events.
Similar MCP servers
All MCP servers for Agent setup: memory, context, orchestration →engram by other authors comes first.
