feat(servers): add synapse-secure-memory (Zero-Knowledge Secure Memory)

[SynapseLayer]

feat(servers): add synapse-secure-memory (Zero-Knowledge Secure Memory)

Server Information

• Name: io.github.synapselayer/synapse-secure-memory
• Title: Synapse Layer — Zero-Knowledge Memory
• Version: 1.0.6
• Transport: stdio
• Package: pypi/synapse-layer

Description

Zero-Knowledge persistent memory for MCP agents with a non-bypassable 4-layer Cognitive Security Pipeline:

1. Semantic Privacy Guard™ — Automatic PII sanitization (12+ regex patterns)
2. Intelligent Intent Validation™ — Auto-categorization + self-healing on recall
3. Differential Privacy — Calibrated Gaussian noise (ε-DP) on embedding vectors
4. Neural Handover™ — HMAC-signed, persistence-first cross-model context transfer

Installation

pip install synapse-layer
MCP Configuration
JSON{
  "mcpServers": {
    "synapse-layer": {
      "command": "uvx",
      "args": ["synapse-layer"]
    }
  }
}
Links

Homepage: https://synapselayer.org/
Documentation: https://synapselayer.org/docs
Repository: https://github.com/SynapseLayer/synapse-layer
PyPI: https://pypi.org/project/synapse-layer/
Technical Deep-Dive: https://dev.to/synapselayer/beyond-context-windows-a-zero-knowledge-memory-reference-implementation-for-the-mcp-ecosystem-4bcg

Checklist

server.json conforms to schema 2025-12-11
Package published on PyPI (v1.0.6)
mcp-name ownership verification tag added to README
Transport: stdio
Environment variables documented
158 tests passing, 95% coverage

[SynapseLayer]

Hi team,

I'm the maintainer of Synapse Layer, an open-source (Apache 2.0) memory SDK for AI agents. I recently opened PR #1129 to register synapse-secure-memory as an MCP server in the Registry.

The server provides two MCP tools (store_memory and recall_memory) that automatically apply PII sanitization (12 patterns covering SSN, CPF, credit cards, API keys, etc.) and ε-differential privacy on embeddings before storage.

This addresses a real gap in the current reference servers: while @modelcontextprotocol/server-memory handles knowledge graph triples, it doesn't include content-level privacy treatment. For teams in regulated environments, this is a compliance concern.

Happy to iterate on the implementation, naming conventions, or metadata format to match the Registry's conventions.

One question: Does the Registry have a preferred pattern for servers that process data locally (zero external API calls)? I'd like to make sure the description accurately reflects the zero-knowledge architecture.

Best regards,
Ismael Marchi
Synapse Layer
founder.synapselayer@proton.me