ThinkMem

ThinkMem is an MCP server designed to help LLMs manage memory efficiently. It exposes two main operation modes: a local stdio mode for embedding ThinkMem into a single AI assistant workflow, and a StreamableHTTP mode for hosting a memory service that multiple AI assistants can connect to over HTTP. ThinkMem supports multiple memory types including RawMemory for unstructured text and ListMemory for structured, ordered data like tasks or workflows. The HTTP mode enables centralized memory management with scalable access via StreamableHTTP, while the stdio mode is simpler and suitable for development and testing. You can interact with the memory system through its command-line interface, and you can integrate memory instances with external AI assistants by registering them with an MCP host using the provided JSON configuration. The documentation also covers memory operations such as writing, updating, and querying memory, as well as memory summarization and search features.

Prerequisites:

• Node.js 18+ and npm installed on your system
• Basic familiarity with MCP concepts (servers, hosts, and transports)

Installation steps:

1. Install ThinkMem globally (as recommended by the project):

npm install -g thinkmem

2. Verify installation:

thinkmem --version

3. Start in HTTP mode (default port 13809) for MCP host integration:

npx -y thinkmem --mode http --port 13809

4. If you prefer the stdio mode for local development, run:

thinkmem --mode stdio

5. If you want to run via MCP configuration, create a config file (as in the mcp_config below) and start the server using the same command approach.

Note: Depending on your environment, you may need to adjust firewall rules to expose the HTTP port and ensure SSL/TLS termination is handled by a reverse proxy for production deployments.

Tips and common considerations:

• In HTTP mode, ensure the port you choose is accessible from your MCP Host and any clients. Consider placing a reverse proxy in front of ThinkMem for TLS termination in production.
• When using the stdio mode, explicitly specify the memory DB path to avoid collisions between multiple processes.
• The configuration supports multiple AI assistants by registering them in the host with the ThinkMem server as a streamable-http endpoint. Example registrations are shown in the README.
• For persistent memory, ThinkMem stores data in JSON files by default; ensure the filesystem has adequate permissions and space, and enable backups if necessary.
• If you run into issues, check the project's tests and documentation links in the repository for troubleshooting and compatibility notes.