Amber

What the Amber MCP server does

Amber MCP server adds long-term memory to compatible AI assistants. It can retain preferences, decisions, project context, and personal details, then make those facts searchable in later conversations. The assistant handles memory extraction, topic assignment, and query phrasing, while Amber stores and retrieves the resulting records.

Conversations are split into atomic facts instead of being saved as one large text block. Each fact can have its own subjects, topics, dates, and searchable representations. This lets an assistant retrieve a relevant detail without loading an entire conversation.

How it works

Memory storage runs asynchronously, so saving information does not block the conversation. When a later query needs context, the assistant can search the service. Search combines vector retrieval with full-text keyword search and Reciprocal Rank Fusion. Natural-language time references such as “last week” or “3 days ago” are resolved to date ranges for temporal searches.

Each fact receives multiple semantic variants, with at least five required for storage. Automatic topic grouping also connects related terms during retrieval; a search for “work” can find memories associated with “career” or “job.” Memory text is sent to an embedding endpoint rather than a generative chat model. The assistant performs chunking, topic creation, and phrasing, and no generative model is shown the stored memories.

Setup and configuration

Amber MCP server is provided as a managed Streamable HTTP service at https://mcp.ambermem.com. Authentication uses automatically discovered OAuth 2.1, and the README documents no environment variables or local configuration requirements.

Claude Code and Claude Desktop can add the endpoint with the claude mcp add command. Cursor, Windsurf, and VS Code configure the same URL in their respective MCP configuration files. Other MCP clients can connect using the endpoint and Streamable HTTP transport. ChatGPT is documented as supporting the endpoint through Settings, Connectors, and Create.

Tools and capabilities

The service exposes 29 MCP tools:

• 14 memory tools for storing and searching persistent context
• 13 account tools
• 2 feedback or notification tools

The documented feedback path includes amber_send_feedback_to_developer. Account features include data export and account deletion, while billing uses PayPal. Stored data can be exported as JSON.

Limitations and notes

Amber MCP server is not currently self-hosted; it is a managed service operated on Cloudflare Workers with Turso databases. Each user receives an isolated database. The service advertises a 90-day free trial followed by a $2.99 monthly charge through PayPal, and PayPal is currently required for identity and billing.

The README states that search adds less than one second and that storage is processed in the background. It also describes GDPR compliance, no email collection, and no marketing messages. Users can cancel through the assistant or PayPal and can export their data before leaving.