# Giskard Memory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/giskard09/giskard-memory  
**GitHub Stars:** 0  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/giskard-memory

## Description
Pay-per-use semantic memory for AI agents. SHA256 commitment, Ed25519 sig, Lightning.

## 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": {
  "giskard-memory": {
    "url": "https://your-tunnel.trycloudflare.com/sse"
  }
}
```

## Documentation

## What Giskard Memory does

The Giskard Memory MCP server provides persistent memory for agents that would otherwise lose their context when a session ends. It accepts text for storage and lets an agent search previously stored information by semantic meaning. Memories are associated with an agent identity, allowing the caller to provide the same identifier when storing and recalling information.

The service uses ChromaDB as its vector database and Sentence Transformers to create semantic embeddings. This supports queries such as asking what an agent knows about a previously discussed project without requiring the query to repeat the original wording.

## How it works

Payment is part of each memory operation. Before saving information, the agent requests an invoice for the `store` action, pays the resulting 5-sat Lightning invoice, and then submits the content, agent ID, and payment hash to `store_memory`. Recalling follows the same pattern: the agent requests a `recall` invoice, pays 3 sats, and calls `recall_memory` with a query, agent ID, and payment hash.

The Giskard Memory MCP server communicates with MCP clients through an HTTP SSE endpoint. The repository example uses a public Cloudflare Tunnel URL that forwards to the locally running service.

## Setup and configuration

Run the service locally by cloning the repository, installing its Python dependencies, and starting `server.py`:

```bash
git clone https://github.com/giskard09/giskard-memory
cd giskard-memory
pip install mcp httpx chromadb sentence-transformers python-dotenv
python3 server.py
```

Create a `.env` file containing the Phoenix daemon password:

```text
PHOENIXD_PASSWORD=your_phoenixd_password
```

The README starts the service on port 8001 and uses Cloudflare Tunnel to expose it publicly:

```bash
cloudflared tunnel --url http://localhost:8001
```

Add the resulting `/sse` URL to the MCP client's server configuration.

## Tools and capabilities

The Giskard Memory MCP server exposes these documented tools:

- `store_memory`: Saves text as a memory for an agent identity.
- `recall_memory`: Retrieves stored memories using a meaning-based query.
- `get_invoice`: Creates a Lightning invoice for either storing or recalling.
- `get_status`: Reports the service name, version, port, uptime, health status, and dependencies.

The documented prices are 5 sats for storing and 3 sats for recalling. The stack combines MCP, ChromaDB, Sentence Transformers, phoenixd, and Cloudflare Tunnel.

## Limitations and notes

Public access also depends on separately running Cloudflare Tunnel. Memory retrieval is semantic rather than exact keyword matching, and every documented store or recall operation requires payment first.

_Full upstream README: https://allmcps.com/mcp/giskard-memory/readme_