Engram

edg-l18stars

MCP server in Rust for AI agent persistent memory: branch-aware session handoffs, local ONNX embeddings, SQLite-backed semantic search.

Knowledge Memory Databases

Install Run on PlugMCP

GitHub Website

Run on PlugMCP

1# Engram

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 18
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Engram?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Engram need an API key?

No required environment variable is listed on this record.

What transport does Engram use?

Engram runs locally over stdio.

When was Engram last checked?

Last checked 2026-10-28. MCP server in Rust for AI agent persistent memory: branch-aware session handoffs, local ONNX embeddings, SQLite-backed semantic search.

From the publisher

Documentation

Everything stays on your machine: one SQLite file and a small local embedding model. Nothing is sent to a cloud service, and storing a memory never calls an LLM.

A session, resumed

$ engram-cli handoff resume
Branch: feat/auth
Latest handoff: mem_a5b0…
Chain (2 handoffs, oldest to newest):
  mem_e8cc…
  mem_a5b0…

Open todos (3 total):
  - [ ] Rotate the signing key before the 2.0 tag (mem_65e6…)
  ...

Open blockers:
  - Staging has no OAuth client for the new callback URL

Top sections:
─────────────────────────────────────────
[mem_a5b0…] tried (score: 0.94)
Refreshing tokens in middleware: every request raced the refresh and logged users out
─────────────────────────────────────────
...

An agent gets the same thing from the handoff_resume tool, or from the /mcp__engram__resume prompt in Claude Code.

Install

cargo install engram_mcp
claude mcp add -s user engram $(which engram)

This installs two binaries. engram is the MCP server, which your client starts. engram-cli is for you and for scripts. For Claude Desktop, the session hooks and the handoff skills, see setup.

What it does

Handoffs are structured into sections and chained per git branch, so context from feat/auth does not leak into fix/billing.
Todos live on a durable per-project list with an explicit done or dropped state, so finished work and forgotten work stay distinguishable.
Facts, decisions, patterns and debug findings are searched with BM25 and embeddings combined. A store reports existing memories it may repeat or replace.
Old knowledge fades as newer knowledge about the same project arrives, not with the calendar. A project you leave for a year is as fresh as the day you left it.
Architecture decisions are kept as numbered ADRs that move through proposed, accepted and superseded, and export as Markdown files.
A memory can be superseded by a newer one, which retrieval then returns in its place. Deletions go to a trash you can restore from.
Project ids come from the git remote, so the same repo is the same project on every machine. engram-cli sync user@host keeps two machines in step over plain ssh.

Documentation


Setup	Clients, hooks, skills, syncing two machines
MCP tools	Every tool, its arguments, project and branch scoping
CLI	`engram-cli` commands and `--json` output
Configuration	Environment variables
How it works	Retrieval, decay, dedup, handoff chains, sync rules, benchmark
Compared to other tools	Mem0, mcp-memory-service, Zep, Letta
Hooks	Passive capture from Claude Code lifecycle events
Changelog

Limits

Single user, single node, no auth: it is a local MCP server. Sync is pairwise between machines you can ssh into, and deletions do not sync.

Development

cargo build --release
cargo test
cargo clippy --all-targets
cargo fmt --check

License

Apache-2.0

Engram

edg-l18stars

MCP server in Rust for AI agent persistent memory: branch-aware session handoffs, local ONNX embeddings, SQLite-backed semantic search.

Knowledge Memory Databases

Install Run on PlugMCP

GitHub Website

Run on PlugMCP

1# Engram

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 18
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Engram?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Engram need an API key?

No required environment variable is listed on this record.

What transport does Engram use?

Engram runs locally over stdio.

When was Engram last checked?

Last checked 2026-10-28. MCP server in Rust for AI agent persistent memory: branch-aware session handoffs, local ONNX embeddings, SQLite-backed semantic search.

From the publisher

Documentation

Everything stays on your machine: one SQLite file and a small local embedding model. Nothing is sent to a cloud service, and storing a memory never calls an LLM.

A session, resumed

$ engram-cli handoff resume
Branch: feat/auth
Latest handoff: mem_a5b0…
Chain (2 handoffs, oldest to newest):
  mem_e8cc…
  mem_a5b0…

Open todos (3 total):
  - [ ] Rotate the signing key before the 2.0 tag (mem_65e6…)
  ...

Open blockers:
  - Staging has no OAuth client for the new callback URL

Top sections:
─────────────────────────────────────────
[mem_a5b0…] tried (score: 0.94)
Refreshing tokens in middleware: every request raced the refresh and logged users out
─────────────────────────────────────────
...

An agent gets the same thing from the handoff_resume tool, or from the /mcp__engram__resume prompt in Claude Code.

Install

cargo install engram_mcp
claude mcp add -s user engram $(which engram)

This installs two binaries. engram is the MCP server, which your client starts. engram-cli is for you and for scripts. For Claude Desktop, the session hooks and the handoff skills, see setup.

What it does

Handoffs are structured into sections and chained per git branch, so context from feat/auth does not leak into fix/billing.
Todos live on a durable per-project list with an explicit done or dropped state, so finished work and forgotten work stay distinguishable.
Facts, decisions, patterns and debug findings are searched with BM25 and embeddings combined. A store reports existing memories it may repeat or replace.
Old knowledge fades as newer knowledge about the same project arrives, not with the calendar. A project you leave for a year is as fresh as the day you left it.
Architecture decisions are kept as numbered ADRs that move through proposed, accepted and superseded, and export as Markdown files.
A memory can be superseded by a newer one, which retrieval then returns in its place. Deletions go to a trash you can restore from.
Project ids come from the git remote, so the same repo is the same project on every machine. engram-cli sync user@host keeps two machines in step over plain ssh.

Documentation


Setup	Clients, hooks, skills, syncing two machines
MCP tools	Every tool, its arguments, project and branch scoping
CLI	`engram-cli` commands and `--json` output
Configuration	Environment variables
How it works	Retrieval, decay, dedup, handoff chains, sync rules, benchmark
Compared to other tools	Mem0, mcp-memory-service, Zep, Letta
Hooks	Passive capture from Claude Code lifecycle events
Changelog

Limits

Single user, single node, no auth: it is a local MCP server. Sync is pairwise between machines you can ssh into, and deletions do not sync.

Development

cargo build --release
cargo test
cargo clippy --all-targets
cargo fmt --check

License

Apache-2.0