# agenticmemory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/jyswee/agenticmemory  
**GitHub Stars:** 1  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/agenticmemory

## Description
Persistent memory for AI agents — history, context, semantic search, queues. 17 tools, stdio/SSE.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "agenticmemory": {
    "url": "https://mcp.agenticmemory.ai/sse"
  }
}
```

## Documentation

## What agenticmemory MCP server does

The agenticmemory MCP server connects MCP-compatible agents to persistent memory stored in Agentic Memory. It separates memory into several forms: ordered conversation history, typed key-value context for durable facts and decisions, titled and tagged entries for longer-lived knowledge, and scratchpad data for temporary work in progress.

Agents can retrieve recent history, load a compact session state, and search older material semantically. The bootstrap operation combines messages, context, and entries into one session-start response, while a semantic bootstrap can focus that response on a topic. Spaces allow multiple agents, projects, or interfaces to use shared memory.

The service also provides FIFO queues within spaces. One agent can push an opaque JSON work item and another can pop it, wait for new work, inspect the queue, or review dead-lettered items. This supports handoffs without adding a separate queue service.

## How it works

The package is installed as the `agmry` command and exposes an MCP process through `agmry mcp-serve`. Local MCP clients launch that process over stdio. The repository also documents a hosted SSE endpoint at `https://mcp.agenticmemory.ai/sse`, which accepts a bearer API key and requires no local installation.

The agenticmemory MCP server exposes 17 native tools covering storage, recall, search, bootstrap, context, entries, spaces, and queues. API access is associated with a project or space and can be shared across the CLI, MCP interface, and REST interface.

## Setup and configuration

Install the CLI globally with `npm install -g agmry`. For a local MCP configuration, set `AGMRY_API_KEY` in the client environment because the MCP process runs outside the project directory and does not automatically read `.agmry/config.json`.

JSON-based clients can configure `agmry` as the command with `mcp-serve` as its argument and pass the API key through `env`. Remote clients use the documented SSE URL and an `Authorization: Bearer` header.

For encrypted spaces, generate or supply an encryption key. Resolution checks command-line flags first, then `AGMRY_ENCRYPTION_KEY`, followed by project-level and home-level `.agmry/config.json` files. A passphrase can be used instead of storing a key.

## Tools and capabilities

The agenticmemory MCP server supports these capability groups:

- Store and recall role-aware conversation messages.
- Set and retrieve durable context values.
- Create, list, and search titled or tagged entries.
- Perform semantic search across remembered information.
- Bootstrap a compact session state, optionally around a topic.
- Create and use shared spaces.
- Push, pop, wait on, inspect, and dead-letter queue items.
- Use managed encryption at rest or client-side end-to-end encryption.

Zero-knowledge spaces encrypt data before it leaves the client. The server cannot read that data, so semantic search is unavailable there. Managed encryption keeps server-side features such as search and summaries available.

## Limitations and notes

Losing an end-to-end encryption key or passphrase makes the associated data unrecoverable. The README recommends backing up the key. Queue envelopes are opaque JSON; the server does not inspect their contents. A queue pop on an empty queue returns exit code 2 in the CLI, while a wait option can long-poll for new work.

The package requires an API key for the documented hosted service configuration. The README names CLI, MCP, REST, and dashboard access, but the MCP listing itself does not describe the underlying API schema or individual tool signatures beyond their capability groups.

_Full upstream README: https://allmcps.com/mcp/agenticmemory/readme_