tiger-memory

Tiger Memory MCP Server provides a simple in-memory memory storage system with tools exposed via the Model Context Protocol (MCP) and REST endpoints. It enables LLMs to store, retrieve, and manage contextual data quickly and securely, making it easier to reference relevant facts during conversations. The server exposes a set of MCP tools that can be discovered and invoked programmatically, and it also offers REST API endpoints for integration with other services. You can deploy the server locally for development or in a Kubernetes cluster for production usage, backed by a Timescale database for persistence when needed.

To use the server locally, start it with Node.js and connect via MCP tools or the REST API. Tools are accessible through your MCP client, inspector, or client stub, and you can inspect the available capabilities using the MCP Inspector (as described in the repo). The README provides an example configuration for Claude Desktop to integrate with the server by pointing to the local dist/index.js and passing the stdio transport. This makes it straightforward to test prompts, store snippets, and retrieve stored memories during a session. The API supports standard CRUD operations for memory slots, plus domain-specific helpers tailored for fast retrieval and lexical matching.

Prerequisites:

• Node.js (version 14+ recommended) and npm installed
• Access to a terminal with git
• Optional: a running PostgreSQL/TimescaleDB instance if you plan to persist data beyond memory

1. Clone the repository

git clone git@github.com:timescale/tiger-memory-mcp-server.git
cd tiger-memory-mcp-server

2. Install dependencies and build

npm i

3. Create a local environment file

cp .env.sample .env

Edit the .env file as needed for local development (variables defined in .env.sample).

4. Run the server (development)

npm run start

Or watch for changes and rebuild

npm run watch

5. Optional: run the MCP Inspector to test endpoints

npm run inspector

6. If you plan to integrate with Claude Desktop or other MCP clients, configure the client to point to the server's index file as shown in the README example (absolute path to dist/index.js) and provide transport settings (e.g., stdio).

Tips and common considerations:

• If you’re using a Kubernetes deployment, you’ll likely want to configure a Helm chart in chart/ as indicated in the README and wire up the database secrets using sealed secrets in your cluster.
• Ensure your database connection details (PGHOST, PGDATABASE, PGPORT, PGUSER, PGPASSWORD) are secured and not committed to version control when using real credentials.
• The server supports REST endpoints in addition to MCP tools; you can test endpoints with curl or your preferred HTTP client.
• The environment variables for database connections are the most common source of runtime errors; double-check hostnames, ports, and credentials when deploying to different environments (dev/stage/prod).
• If you encounter issues with the inspector or adapter tooling, verify that the dist/index.js path used by MCP clients matches the built output of your local build (npm run build or npm run watch).