Skip to content

MCP Server

Connect Engram to any MCP-compatible AI agent

MCP Server

Connect Engram Memory to any AI agent that supports the Model Context Protocol (MCP).

What is MCP?

The Model Context Protocol is an open standard for connecting AI models to external tools and data sources. Instead of custom API integrations, agents discover and call tools through a unified protocol.

Engram ships an MCP server that exposes memory operations as tools. Any MCP-compatible client (Claude Code, Claude Desktop, OpenClaw, Cursor, Windsurf, etc.) can store, search, and manage memories without writing integration code.

MCP Tools

The Engram MCP server exposes 6 tools:

ToolDescription
memory_storeStore a new memory with optional category and metadata
memory_searchSemantic search across stored memories
memory_recallRecall relevant memories for a given context (alias for search)
memory_forgetDelete memories by ID or query match
memory_consolidateMerge duplicates and optimize storage
memory_connectHealth check and connection status

Transport Types

The server supports multiple transport protocols:

TransportURLUse Case
Streamable HTTPhttp://localhost:8585/mcpRecommended default. Works with most clients.
SSEhttp://localhost:8585/mcp/sseLegacy streaming transport
StdioDirect process spawnLocal-only, no network required
RESThttp://localhost:8585/v1/*Standard REST API fallback

Claude Code

Add Engram as an MCP tool source in one command:

claude mcp add engram-memory --transport http --url http://localhost:8585/mcp

For the cloud-hosted platform:

claude mcp add engram-memory --transport http \
  --url https://api.engrammemory.ai/v1/mcp \
  --header "Authorization: Bearer pr_live_xxxxx"

Verify the connection:

claude mcp list

Once added, Claude Code can use memory_store, memory_search, memory_recall, memory_forget, memory_consolidate, and memory_connect as tools in any conversation.

Claude Desktop

Add the following to your Claude Desktop MCP configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json

Local (self-hosted)

{
  "mcpServers": {
    "engram-memory": {
      "transport": "http",
      "url": "http://localhost:8585/mcp"
    }
  }
}

Cloud (platform)

{
  "mcpServers": {
    "engram-memory": {
      "transport": "http",
      "url": "https://api.engrammemory.ai/v1/mcp",
      "headers": {
        "Authorization": "Bearer pr_live_xxxxx"
      }
    }
  }
}

Restart Claude Desktop after editing the config file.

OpenClaw

OpenClaw is the open-source MCP client from the Engram community.

git clone https://github.com/EngramMemory/openclaw.git
cd openclaw
./install.sh

OpenClaw auto-discovers a local Engram MCP server on http://localhost:8585/mcp. To point it at the cloud platform, set:

export ENGRAM_API_KEY="pr_live_xxxxx"
export ENGRAM_BASE_URL="https://api.engrammemory.ai/v1"

Multi-Client Install

For MCP clients that support the npx install-mcp convention:

npx install-mcp engrammemory

This registers the Engram MCP server with the active MCP client (Claude Desktop, Cursor, Windsurf, etc.) using its standard config path.

Docker

Run the full Engram Community Edition stack in a single container:

docker run -d \
  --name engram \
  -p 8585:8585 \
  -p 6333:6333 \
  -p 11435:11435 \
  -v engram-data:/data \
  engrammemory/engram-stack
PortService
8585MCP server + REST API
6333Vector database
11435Embedding service

Once running, connect any MCP client to http://localhost:8585/mcp.

For full Docker configuration options, see the Self-Hosted Docker Guide.

Verify Connection

From any MCP client, call the memory_connect tool. A successful response looks like:

{
  "connected": true,
  "latency_ms": 12,
  "vector_db": "healthy",
  "embedding": "healthy",
  "version": "0.2.0"
}

You can also verify with curl:

curl http://localhost:8585/v1/connect

Next Steps