Synap Memory

What Synap Memory MCP server does

The Synap Memory MCP server gives an MCP-compatible agent access to persistent memory backed by Synap Cloud. It can record messages from a conversation and later retrieve context relevant to a user or query. The memory can remain available across separate conversations, sessions, and devices, rather than being limited to the current interaction.

The underlying SDK is available for Python and JavaScript/TypeScript, while the repository also includes an MCP adapter. The service is intended for agents that need to retain facts, preferences, or ongoing work between interactions. A typical flow records a user's message in one conversation, then fetches related context in a later conversation.

How it works

Memory processing runs in Synap Cloud. The package is not a local memory engine, and the quickstart does not require running a local server. Requests are authenticated with a Synap API key and sent to the hosted service.

Synap's documented memory pipeline includes anticipatory retrieval, entity resolution, temporal relevance, contradiction handling, custom memory architectures, and hierarchical scoping. The scoping model covers client, customer, user, and conversation levels. A fetch operation can merge relevant information from these levels, which supports both multi-tenant applications and per-user context.

The README reports approximately 15 ms P50 context-fetch latency from an internal measurement. It also describes benchmark results of 92.0% on LongMemEval and 93.2% on LoCoMo, using the project's stated evaluation setup. These figures are reported by the project and should be interpreted within that methodology.

Setup and configuration

Install the Python package with pip install maximem-synap, or install the JavaScript package with npm install @maximem/synap-js-sdk. For direct SDK usage, create a client with an API key and initialize it before recording or fetching memory.

For MCP use, add the Synap MCP server URL and an API key obtained from the Synap dashboard to an MCP client. The repository points Cursor users and other MCP clients to the MCP server package for configuration details. Claude Code and Codex also have repository-provided skills that can help wire Synap into an agent project.

No environment-variable names are specified in the provided material, so configuration should not be assumed to use a particular variable. The required credential is described as a Synap API key.

Tools and capabilities

The documented operations are:

• Record a message with a conversation ID, user ID, role, and content.
• Fetch context using a user or conversation scope and a search query.
• Return formatted context for insertion into an agent workflow.
• Resolve references across sessions, such as connecting a person's name with an earlier description.
• Apply time-aware relevance and process explicit retractions or contradictions.
• Scope memory across client, customer, user, and conversation levels.

The Synap Memory MCP server is a fit when an agent must recall information after the original conversation has ended. It is less suitable if the requirement is a fully local memory service, because the documented engine runs in Synap Cloud and requires an API key.