Mem-graph Memory MCP Server

Mem-graph Memory is an agent memory store for Mem-graph implemented as an MCP (Model Context Protocol) server. It leverages the FastMCP framework to provide robust capability for capturing and inter-linking conversations, tasks, decisions, notes, and audit violations, enabling semantic recall across AI assistant sessions.

Features & Capabilities

The server provides a suite of MCP tools categorized by domain:

1. Memory Management

Tools to capture arbitrary information and retrieve it semantically across sessions.

• memory_store: Store a specific memory or information snippet dynamically.
• memory_recall: Recall memories by querying specific concepts or topics.
• memory_search: Perform semantic search over all stored memories.
• memory_list: List stored memories.
• memory_expire: Expire or remove a memory when it's no longer relevant.

2. Conversational Tracking

End-to-end conversation transcript storage and summarization.

• conversation_start: Initiate the tracking of a new conversation session.
• conversation_append: Append transcript data or messages to the ongoing conversation.
• conversation_end: End a conversation and automatically generate summaries.
• conversation_get: Retrieve the details and transcript of a specified conversation.

3. Project Management

Tools to define and track larger overarching projects.

• project_create: Initialize a new project.
• project_get: Retrieve a project by its identifier.
• project_list: List active and inactive projects.
• project_search: Search through the project repository.

4. Task Tracking

Fine-grained task definition, updates, and linking.

• task_create: Create a new task within a project.
• task_update: Update an existing task's status, assignee, or details.
• task_get: Fetch the current state of a task.
• task_search: Look up tasks matching specific criteria.
• task_link_decision: Link a task to an architectural or structural decision.
• task_link_violation: Link a task to an identified rule or audit violation.
• task_block: Mark a task as blocked and optionally record the reason.

5. Architectural Decisions

Formal tracking of decisions that impact the codebase or project trajectory.

• decision_record: Record a new decision, rationale, and context.
• decision_supersede: Mark an older decision as superseded by a newer one.
• decision_get: Retrieve the details of a specific decision.
• decision_search: Search historical decisions.

6. Notes

Ad-hoc text and documentation storage.

• note_create: Create a freeform note.
• note_search: Search existing notes.
• note_list: List all notes.

7. Violations & Auditing

Tools to identify, record, and resolve rule violations or bad practices.

• violation_record: Record an observed code, architecture, or workflow violation.
• violation_resolve: Mark a documented violation as resolved.
• violation_recur: Log when a resolved violation recurs.
• violation_search: Search through the database of recorded violations.
• violation_list: List accumulated violations to track frequency or severity.

Architecture

The MCP Server is built using:

• FastMCP: Provides the foundation for routing, lifecycle, and multiple transport supports (stdio, chunked streamable HTTP, and SSE).
• Ladybug DB: Serves as the underlying robust graph database where all these entities are interlinked and serialized to facilitate semantic querying across nodes.
• Ollama: Generates local, dense semantic embeddings of textual data enabling nearest-neighbor concept searches natively across your tracked interactions and states.