# Engram [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/get-engram/engram  
**GitHub Stars:** 19  
**npm Downloads (last month):** 291  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/engram-2

## Description
Persistent memory for AI agents — verbatim conversations, searchable by meaning.

## Tools
Capabilities this server exposes over MCP:

- **create_conversation** — Start a new conversation with optional title, tags, metadata
- **append_messages** — Add messages to an existing conversation
- **search** — Semantic search across all conversations
- **get_conversation** — Retrieve a conversation with its messages
- **list_conversations** — List conversations with filtering and pagination
- **delete_conversation** — Remove a conversation and its data

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "engram": {
    "command": "npx",
    "args": ["-y","@getengram/cli"]
  }
}
```

## Documentation

## What the Engram MCP server does

The Engram MCP server provides persistent storage for AI agent conversations. It keeps messages verbatim rather than summarizing or compressing them, while also indexing their meaning for semantic retrieval. Conversation records can include titles, tags, and metadata, giving agents additional ways to organize stored context.

This fits agents that need continuity across sessions, such as assistants that should recall earlier discussions or workflows that need to revisit prior conversation records. Storage is isolated by organization, and the service supports team seats and API key management according to the project documentation.

## How it works

The Engram MCP server runs as a hosted service at `https://mcp.getengram.app/mcp`. Its backend uses Cloudflare Workers for the API and MCP implementation, D1 for message and metadata storage, Vectorize for the semantic index, and Workers AI to generate embeddings. The documented embedding model is `bge-base-en-v1.5`.

Clients authenticate through the MCP authorization flow, including browser sign-in and PKCE. The quick-start instructions state that users do not need to copy an API key or create a local configuration file when using the supported OAuth flow. A free account is created during sign-in.

## Setup and configuration

For HTTP-capable MCP clients, add the hosted endpoint as an HTTP server:

```json
{
  "mcpServers": {
    "engram": {
      "type": "http",
      "url": "https://mcp.getengram.app/mcp"
    }
  }
}
```

Clients that only support stdio can use `mcp-remote` with the same endpoint. That bridge starts through `npx` and opens the browser-based OAuth flow. The repository specifically notes that the older `npx @getengram/cli mcp` configuration does not work, so it should not be used for setup.

Claude Code can install the provided plugin through its marketplace commands. The plugin also includes a skill that guides Claude on when to search memory and what information may be worth saving.

## Tools and capabilities

The Engram MCP server exposes six tools:

- `create_conversation` starts a conversation and accepts an optional title, tags, and metadata.
- `append_messages` adds messages to an existing conversation.
- `search` performs semantic search across stored conversations.
- `get_conversation` returns a conversation together with its messages.
- `list_conversations` lists records with filtering and pagination.
- `delete_conversation` removes a conversation and its data.

Together, these operations cover the basic lifecycle of stored conversations: create a record, add transcript messages, find relevant context, inspect records, browse them, and remove them when necessary. The search operation is meaning-based rather than limited to exact text matching.

## Limitations and notes

Usage is subject to plan limits: the listed Free plan includes 1,000 messages per month, while paid plans provide higher quotas. The repository is licensed under Business Source License 1.1.

The README names Claude Desktop, Claude Code, Cursor, Windsurf, Zed, and OpenAI Codex CLI as supported integration targets. More generally, it describes the service as compatible with MCP clients, with `mcp-remote` available when a client lacks stdio support for the hosted connection.

_Full upstream README: https://allmcps.com/mcp/engram-2/readme_